Merge branch 'master' into feat/produced-files-folder
This commit is contained in:
Ziya
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-06-15-code-mode.md
|
||||
2026-06-15-code-mode.md: 51c53c56f5755d56f49d8e5166a1ceeef0ffc202
|
||||
2026-06-15-code-mode.md: f9bde7a85d9cf26d3d6670744c955380fa807e3b
|
||||
2026-06-15-code-mode.zh.md: dbf8d409dea152b39d215f5dd636989c3f4fa0bb
|
||||
|
||||
@@ -89,7 +89,7 @@ The SDK instructs the model to write an async body in the loaded runtime's langu
|
||||
|
||||
## Consequences
|
||||
|
||||
Deployments switching to `'code'` must update any native-only `toolOrder`. Assembly listeners own the integrity of any rewritten protocol surface. Sub-dispatch starts in submission order under a bounded overlap pool, while per-call contexts retain their source, envelope, and metadata through the outer result.
|
||||
Deployments switching to `'code'` must update any native-only `toolOrder`. Assembly listeners own the integrity of any rewritten protocol messages. Sub-dispatch starts in submission order under a bounded overlap pool, while per-call contexts retain their source, envelope, and metadata through the outer result.
|
||||
|
||||
## Testing
|
||||
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-06-17-filesystem-tool-schemas.md
|
||||
2026-06-17-filesystem-tool-schemas.md: ede7daee5c0f49a094df16861b107c8f528664fa
|
||||
2026-06-17-filesystem-tool-schemas.md: 5fff2c8781ff462265a9cb2042aca2522da87b04
|
||||
2026-06-17-filesystem-tool-schemas.zh.md: 5ab19b82d6523bc81dd57c319252d998cc203911
|
||||
|
||||
@@ -6,7 +6,7 @@ English | [中文](2026-06-17-filesystem-tool-schemas.zh.md)
|
||||
|
||||
## Problem
|
||||
|
||||
[The filesystem capability-seam Agent Note](../architecture/2026-06-17-filesystem-capability-seam.md) defines the filesystem capability seam (`ctx.fs`), the package split (`dsh-fs`, `dsh-fs-local`, `dsh-tool-fs`, plus the `dsh-fs-policy` policy plugin), and the observed-file/stale-version policy for read-before-write/edit checks — which the [split-fs-seam](../simplification/2026-06-26-fsspec-style-fs-seam.md) and [event-gate](../architecture/2026-06-26-file-context-as-event-gate.md) Agent Notes moved off `ctx.fs` into the `dsh-fs-policy` plugin on the `fs/*` event gate. The remaining decision for the first filesystem tool delivery is the model-facing schema surface: what arguments the model sees for `read`, `write`, and `edit`.
|
||||
[The filesystem capability-seam Agent Note](../architecture/2026-06-17-filesystem-capability-seam.md) defines the filesystem capability seam (`ctx.fs`), the package split (`dsh-fs`, `dsh-fs-local`, `dsh-tool-fs`, plus the `dsh-fs-policy` policy plugin), and the observed-file/stale-version policy for read-before-write/edit checks — which the [split-fs-seam](../simplification/2026-06-26-fsspec-style-fs-seam.md) and [event-gate](../architecture/2026-06-26-file-context-as-event-gate.md) Agent Notes moved off `ctx.fs` into the `dsh-fs-policy` plugin on the `fs/*` event gate. The remaining decision for the first filesystem tool delivery is the model-facing schema: what arguments the model sees for `read`, `write`, and `edit`.
|
||||
|
||||
The schema must be small, yet stable enough that local/remote/sandboxed filesystem backends do not require model-facing churn, and must avoid importing every option from reference systems. Claude Code and OpenCode expose similar core file tools but differ in naming style and extra flags; this decision picks the minimal shared surface.
|
||||
|
||||
@@ -100,7 +100,7 @@ Schema tests pin the required/optional argument set per tool, empty-`old_string`
|
||||
## Alternatives considered
|
||||
|
||||
- **A Codex-style patch grammar or multi-mode edit API** — rejected: one strict literal replacement mode keeps the model-facing contract simple and lets the backend own exact-match, duplicate-match, line-ending, and stale-version semantics.
|
||||
- **camelCase argument names (OpenCode's style)** — snake_case aligns with Claude Code and the existing harness tool-schema examples, and naming is public surface once shipped.
|
||||
- **camelCase argument names (OpenCode's style)** — snake_case aligns with Claude Code and the existing harness tool-schema examples, and naming is public API once shipped.
|
||||
- **Model-facing `expected_hash` / `expected_version` / `create_only` parameters** — rejected: stale checks are driven by backend-minted versions and the policy plugin's observed state, never by fragile model-copied tokens.
|
||||
|
||||
## Consequences
|
||||
@@ -109,4 +109,4 @@ Schema tests pin the required/optional argument set per tool, empty-`old_string`
|
||||
|
||||
**No explicit model-facing stale guard in v1.** The schema does not ask the model to provide an expected hash/version. That is intentional: stale checks come from backend-produced versions and the `dsh-fs-policy` plugin's observed state, not from fragile model-copied tokens. Filesystem safety failures surface through structured `FsError` codes owned by `dsh-fs`, not through model-supplied version fields.
|
||||
|
||||
**Naming becomes public surface.** Once shipped, changing `file_path` to `filePath` or `old_string` to `oldString` would churn prompts, examples, and downstream clients. This Agent Note chooses snake_case up front and treats it as the stable model-facing contract.
|
||||
**Naming becomes public API.** Once shipped, changing `file_path` to `filePath` or `old_string` to `oldString` would churn prompts, examples, and downstream clients. This Agent Note chooses snake_case up front and treats it as the stable model-facing contract.
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-06-18-compaction-capability-seam.md
|
||||
2026-06-18-compaction-capability-seam.md: 83f224aef50eb02712ac095da0cabf552399a9a6
|
||||
2026-06-18-compaction-capability-seam.md: 2c8128b187f049b584af32f4f40f8e72da56dd73
|
||||
2026-06-18-compaction-capability-seam.zh.md: 12cf2498535d77dcf1a718b841a751bccadfdf1c
|
||||
|
||||
@@ -16,7 +16,7 @@ Two forces shape the design. First, compaction policy and reusable token measure
|
||||
|
||||
### Compaction is a capability seam with separate Service Definition and Service provider roles
|
||||
|
||||
Per the [capability-seams Agent Note](../architecture/2026-06-13-capability-seams.md), compaction ships as separate packages so the contract, the algorithm, and (later) the consumer surface evolve independently:
|
||||
Per the [capability-seams Agent Note](../architecture/2026-06-13-capability-seams.md), compaction ships as separate packages so the contract, the algorithm, and (later) the consumer API evolve independently:
|
||||
|
||||
1. **Interface** — `@deepseek-ai/dsh-compact`: an abstract `CompactService` owning the `ctx.compact` key, the `CompactionResult` vocabulary, the `compact/*` session events, the manual failure taxonomy, and the canonical checkpoint message source. It declares `compactIfNeeded()`, `compactNow()`, and `compactRegion()` as **abstract** — the contract states *what* compaction does, not *how*.
|
||||
2. **Implementation** — `@deepseek-ai/dsh-compact-basic`: a concrete `BasicCompactService` that consumes `ctx.tokenMeter` and owns the tail→head retention walk, summarization via `ctx.llm.stream()`, the surface replacement, the lock, pre-step pressure, and canonical context-overflow recovery. `summarize()` is its sole subclass hook; pricing and replay stay with the meter.
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-06-30-hook-protocol-lib.md
|
||||
2026-06-30-hook-protocol-lib.md: e6938f88b2b587ef97c135d6482ae9ce0e3a94c1
|
||||
2026-06-30-hook-protocol-lib.md: ca7d1acd356b71824a9bbfe3857b3c48c5cdff4b
|
||||
2026-06-30-hook-protocol-lib.zh.md: 4a4efae9a8950368d4db0309bc2c89a3f8c6d0d0
|
||||
|
||||
@@ -16,7 +16,7 @@ A new `packages/hooks/` group with `hook-protocol` as a pure library. It owns fo
|
||||
|
||||
**Shared (here):**
|
||||
- **Matcher** — `matcherDiagnostic(pattern, mode)` and `matchesMatcher(pattern, query, mode)`. The ONE axis the dialects differ on is collapsed to the `mode` parameter: `claude` treats a pure `[A-Za-z0-9_|]+` pattern as a literal (pipe = exact-match alternation) and anything else as a regex; `codex` is always an unanchored regex. Match-all on absent/`''`/`'*'`. Each bridge ignores unsupported events before group parsing, discards matcher fields on supported events without matcher subjects, validates the remaining runnable groups, and treats an invalid regex there as a whole-config load failure, with a stable dialect/pattern/event diagnostic; no hook listeners are registered. Runtime matching still contains an invalid regex as a non-match, so a direct library caller never throws into the loop.
|
||||
- **Execution** — `runHook(bash, hook, options)`. Runs a command hook through the `ctx.bash` seam rather than a bespoke `spawn`: the executor already provides the scrubbed-but-overridable env, process-group kills, and timeout the protocol needs, and `dsh-bash`'s `stdin`/`env` fields (added for exactly this) are the trusted-plugin surface an in-process bridge is allowed to use. It serializes the bridge-built payload to stdin (trailing newline iff CC), honors the hook's `timeoutSec` (else `DEFAULT_HOOK_TIMEOUT_MS`, the 10-minute reference default both dialects share), and never throws (an executor rejection becomes a non-blocking-error `HookOutput`).
|
||||
- **Execution** — `runHook(bash, hook, options)`. Runs a command hook through the `ctx.bash` seam rather than a bespoke `spawn`: the executor already provides the scrubbed-but-overridable env, process-group kills, and timeout the protocol needs, and `dsh-bash`'s `stdin`/`env` fields (added for exactly this) are the trusted-plugin API an in-process bridge is allowed to use. It serializes the bridge-built payload to stdin (trailing newline iff CC), honors the hook's `timeoutSec` (else `DEFAULT_HOOK_TIMEOUT_MS`, the 10-minute reference default both dialects share), and never throws (an executor rejection becomes a non-blocking-error `HookOutput`).
|
||||
- **Decode** — `parseHookOutput(exit, stdout, stderr)`, the exit-code + structured-stdout codec, producing a dialect-neutral `HookOutput`. Exit `0` → lenient JSON parse of stdout; exit `2` → blocking error with `stderr` as the reason (surfaced as `decision: 'block'` so no caller needs a separate exit-code branch); other → non-blocking error. Parses the CC structured-stdout fields that have a consumer on some path (`continue`/`stopReason`/`decision`/`hookSpecificOutput.{permissionDecision,additionalContext,updatedInput}`/`systemMessage`); the bridge honors only the subset meaningful for its dialect. Fields with no consumer on any path are not parsed at all (CC's `suppressOutput` — hook stdout never enters a transcript here, so there is nothing to suppress; see [the tighten-hook-protocol-contract Agent Note](../simplification/2026-07-04-tighten-hook-protocol-contract.md)).
|
||||
- **Merge** — `mergeHookOutputs(outputs)`, folding multiple matched hooks into one most-restrictive `MergedHookOutcome`: permission precedence **deny > ask > allow**, halt sticky on the first `continue:false`, block reasons joined `\n\n`, context/system-messages accumulated in order.
|
||||
- **`hook/*` session events** — `hook/invoked` / `hook/result`, declaration-merged into `SessionEventMap` (log-only, like `compact/*` — NOT `SurfaceEventType`s), with `appendHookInvoked`/`appendHookResult` helpers so the invoked/result pairing and owner-defined execution relation stay consistent across bridges. `appendHookResult` also owns the durable record's semantics — the decision string (the hook's parsed decision, else `'stop'` on `continue:false`, else `'pass'`) and the 500-character `stderrSummary` truncation derive from the `HookOutput` here, not per-bridge.
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-06-30-interception-extension-points.md
|
||||
2026-06-30-interception-extension-points.md: 68be3c94f55ae13155e9b22cb91ef1ad10b1e0ed
|
||||
2026-06-30-interception-extension-points.md: b6e7e8c94bfff8f483dc9faf7b195b130a7e9953
|
||||
2026-06-30-interception-extension-points.zh.md: 262793dd7a297f343391a22fb8d9f1d57a9ff531
|
||||
|
||||
@@ -6,7 +6,7 @@ English | [中文](2026-06-30-interception-extension-points.zh.md)
|
||||
|
||||
## Problem
|
||||
|
||||
The harness needs a hooks subsystem: users extend or gate the agent at lifecycle points the way Claude Code (CC) and Codex do. The key reframe driving this design is that **"native hooks" are not a package** — a native hook is just an ordinary Cordis plugin subscribing to the canonical lifecycle events. So the real product is a *powerful, well-typed canonical event surface*; the CC/Codex bridges (the `dsh-hooks-claude` / `dsh-hooks-codex` packages) are merely translators that map an external shell-hook protocol onto that same surface. Anything a bridge can do, a plain plugin can do directly — more powerfully (no serialization boundary, full `ctx`, typed returns).
|
||||
The harness needs a hooks subsystem: users extend or gate the agent at lifecycle points the way Claude Code (CC) and Codex do. The key reframe driving this design is that **"native hooks" are not a package** — a native hook is just an ordinary Cordis plugin subscribing to the canonical lifecycle events. So the real product is a *powerful, well-typed canonical event API*; the CC/Codex bridges (the `dsh-hooks-claude` / `dsh-hooks-codex` packages) are merely translators that map an external shell-hook protocol onto that same API. Anything a bridge can do, a plain plugin can do directly — more powerfully (no serialization boundary, full `ctx`, typed returns).
|
||||
|
||||
The surface needs distinct contracts for per-prompt policy (CC's `UserPromptSubmit`), session-start observation (CC's `SessionStart`), pre-tool policy, around-dispatch control, post-tool transformation, final-result observation, and continuation with a model-facing reason. Conflating those phases gives plugins mutation channels they do not need and makes finality depend on listener ordering. The [event-domain-semantics Agent Note](../architecture/2026-06-30-event-domain-semantics.md) supplies the three-domain rule and the typed-Decision idiom; this Agent Note applies them to the lifecycle extension points.
|
||||
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-06-30-session-store-fork-api.md
|
||||
2026-06-30-session-store-fork-api.md: 85e748ba4e42919eac49cae583a08c72e0c3c0a0
|
||||
2026-06-30-session-store-fork-api.md: b634d3e561aa1142b7db1bed30158306a2318060
|
||||
2026-06-30-session-store-fork-api.zh.md: f985c93085cbc2dbb1b8a768dac82a40025ea54d
|
||||
|
||||
@@ -38,12 +38,12 @@ The Host creates the child through the agent registry with the selected seed and
|
||||
|
||||
**Separate `ctx.sessionFork` service.** An earlier iteration shipped this as a separate service; it overfit the capability-seam pattern. The code had no swappable backend, no extra event surface, no independent ownership lifecycle, and no durable behavior beyond `ctx.sessions.create({ seed, meta })`. Keeping a separate package would make callers discover and install a second service just to perform policy around a session-store primitive.
|
||||
|
||||
**Two functions: `snapshot()` plus `fork()`.** This preserved a reusable seed/metadata computation, but the only supported consumer created a session immediately. It also made the surface feel more abstract than the concrete operation users need. A single `fork()` with an explicit `boundary` keeps the API direct while still supporting previous-point forks.
|
||||
**Two functions: `snapshot()` plus `fork()`.** This preserved a reusable seed/metadata computation, but the only supported consumer created a session immediately. It also made the API feel more abstract than the concrete operation users need. A single `fork()` with an explicit `boundary` keeps the API direct while still supporting previous-point forks.
|
||||
|
||||
**Silently clip open turns to the last completed boundary.** That is correct for `dsh-subagent-fork`, where delegation often starts while the parent turn is open and the child should inherit only the completed prefix. It is wrong for ordinary user/session branching because it hides that the requested fork point was not actually a valid boundary and silently drops the parent turn tail.
|
||||
|
||||
## Consequences
|
||||
|
||||
The public surface stays small and discoverable: live session branching is part of `ctx.sessions`, next to `create({ seed })`, rather than a standalone service or a two-step helper pair. Persistence continues to work through existing `session/created` and `session/flush` behavior: a forked child starts life with seeded events, so existing backends persist that seed once and preserve `parentSession` / `seedLength` in the header.
|
||||
The public API stays small and discoverable: live session branching is part of `ctx.sessions`, next to `create({ seed })`, rather than a standalone service or a two-step helper pair. Persistence continues to work through existing `session/created` and `session/flush` behavior: a forked child starts life with seeded events, so existing backends persist that seed once and preserve `parentSession` / `seedLength` in the header.
|
||||
|
||||
The v1 scope still excludes ACP `session/fork`, unloaded persisted-session forking, model-facing tools, and subagent refactors. If a future ACP method is added, it should advertise the capability only after it has protocol and snapshot coverage; this Agent Note adds no ACP wire behavior, so no ACP snapshot is required. Fork-child replay remains covered by the existing [seed-boundary testing Agent Note](../testing/2026-06-22-fork-child-replay-seed-boundary.md); focused store, Host, carrier, and client tests pin the boundary and reconciliation contracts, while the real Chromium scenario pins the assembled message action and lineage tree.
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-05-dynamic-workflows.md
|
||||
2026-07-05-dynamic-workflows.md: 3e491478286eb77b56872fcbbdd5ebb6b62a5545
|
||||
2026-07-05-dynamic-workflows.md: 4679ce47b12fd96d5f8da2e8edad16fde81f4dae
|
||||
2026-07-05-dynamic-workflows.zh.md: 7888d83f981a96ac5eb31d5ca6f1f8d0b4930ec7
|
||||
|
||||
@@ -26,7 +26,7 @@ One deliberate strictness DIVERGENCE from CC: hook misuse — unknown or deferre
|
||||
|
||||
**Trust premise**: workflow scripts have the same trust as the model's bash access. The engine contains buggy scripts and guarantees settled results, JSON-safe values, and cancellation quiescence; it does not defend against hostile code. A vm context and worker thread are not security boundaries: a script can escape to Node APIs with process-wide authority. Sandboxing requires a separate-process or isolated-vm engine behind this seam.
|
||||
|
||||
**Why `node:worker_threads`**: each run gets one unpooled worker. A vm context limits the documented script surface, while message-port RPC bridges `agent()` to host-side child loops. The worker prevents synchronous script work from blocking the host, provides a serialization boundary, and permits forced termination after cancellation. `isolated-vm` was rejected because of its maintenance state and deployment requirements.
|
||||
**Why `node:worker_threads`**: each run gets one unpooled worker. A vm context limits the documented script API, while message-port RPC bridges `agent()` to host-side child loops. The worker prevents synchronous script work from blocking the host, provides a serialization boundary, and permits forced termination after cancellation. `isolated-vm` was rejected because of its maintenance state and deployment requirements.
|
||||
|
||||
The host validates metadata and parses the body before publication. Private enum-keyed payload maps define the wire protocol; pending starts, published child records, one cancellation signal, worker-death reaping, result precedence, and disposal quiescence preserve the subagent run contract across it. The [agent-scope runtime-design Agent Note](../architecture/2026-07-12-agent-scope-runtime-design.md#workflow-children-are-pending-starts-or-published-records) owns those race algorithms.
|
||||
|
||||
@@ -56,7 +56,7 @@ Worker-side logic runs through an in-process `MessageChannel` so V8 coverage mea
|
||||
|
||||
- **Background collection** (start tool → run id → completion notice → collect), designed alongside bash/subagent background unification.
|
||||
- **Journaling + resume** (`resumeFromRunId`, cached agent() prefixes) — implementing it reintroduces CC's determinism bans as a script-contract tightening (scripts may read the clock today).
|
||||
- **Saved/bundled workflows** (a `.deepseek/workflows/` registry, slash-command surface) and **script persistence to a run directory** (the tool-call event already records the script durably).
|
||||
- **Saved/bundled workflows** (a `.deepseek/workflows/` registry, slash-command API) and **script persistence to a run directory** (the tool-call event already records the script durably).
|
||||
- **Nested `workflow()`**, **token `budget`**, and the `effort`/`isolation`/`agentType` agent options (each rejects loud with a message naming it deferred).
|
||||
- **An overall run wall-clock timeout** — cancellation always frees the caller (result settles within the grace), so a cap on total run time is a policy knob for the background redesign, not a correctness need here.
|
||||
- **Engine hardening beyond worker threads**: an isolated-vm or separate-process engine behind the same seam (actual sandboxing; memory limits).
|
||||
@@ -66,7 +66,7 @@ Worker-side logic runs through an in-process `MessageChannel` so V8 coverage mea
|
||||
## Alternatives considered
|
||||
|
||||
- **Hostile-value containment in the host** (trap-free proxy rejection, accessor-never-invoked descriptor walks, realm-side pre-rendering of thrown values, realm-built promises/arrays/error clones with structural fatal recognition): rejected because every defense targets an author the trust premise accepts, while the thread's serialization boundary already makes cross-realm values total by construction.
|
||||
- **In-process `node:vm` execution**: mechanically simplest — no RPC, no thread — but `start()` blocks the caller for the script's initial synchronous slice, a synchronous spin past the first await cannot be killed in-process (the vm `timeout` covers only that first slice), and `dispose()` could only abandon an unsettling script on the host loop. The worker-thread engine keeps the same vm-context script surface while unblocking the host and making termination real.
|
||||
- **In-process `node:vm` execution**: mechanically simplest — no RPC, no thread — but `start()` blocks the caller for the script's initial synchronous slice, a synchronous spin past the first await cannot be killed in-process (the vm `timeout` covers only that first slice), and `dispose()` could only abandon an unsettling script on the host loop. The worker-thread engine keeps the same vm-context script API while unblocking the host and making termination real.
|
||||
- **Background execution as the default** (CC's shape): deferred; foreground-synchronous matches `dsh-tool-subagent`'s cut, and background semantics should be designed ONCE across bash/subagent/workflow rather than per-tool.
|
||||
- **Workflow-layer JSON parsing for `agent({schema})`**: duplicating a seam concern at one consumer while the seam's capability flag stayed dishonestly `false`.
|
||||
- **Meta embedded in the script as `export const meta = {...}`** (CC's exact format): keeps scripts self-contained and CC scripts drop-in, but obtaining meta requires evaluating model-written text on the host. Even an empty timed vm context cannot bound script-controlled getters when the host reads the resulting object. A JSON parameter removes the scanner, evaluation, and host-spin hole; the cost is that a CC script's meta header must move into the parameter (the body stays drop-in).
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-06-approval-seam.md
|
||||
2026-07-06-approval-seam.md: 8aa9986139dae77e08c166b72545bfa688a389e0
|
||||
2026-07-06-approval-seam.md: 46d3186e226f344ee1239dcbdfb6f286b88d84ce
|
||||
2026-07-06-approval-seam.zh.md: ef4ccf5fd2b54888a648737866ff6f5fe1678882
|
||||
|
||||
@@ -93,7 +93,7 @@ Snapshots record allowed and rejected sandbox escalation through `session/reques
|
||||
|
||||
## Alternatives considered
|
||||
|
||||
- **A single registered provider instead of waterfall listeners** — rejected: a `registerProvider()` surface forces every composition question — allowlist pre-filters, external hook deciders, scripted test answers, a policy gate in front of a human — inside one provider implementation. The waterfall gets composition, fail-closed absence, and HMR disposal from machinery the runtime already has; the seam's JSDoc pins the single-decision-slot convention instead of inventing a provider registry.
|
||||
- **A single registered provider instead of waterfall listeners** — rejected: a `registerProvider()` API forces every composition question — allowlist pre-filters, external hook deciders, scripted test answers, a policy gate in front of a human — inside one provider implementation. The waterfall gets composition, fail-closed absence, and HMR disposal from machinery the runtime already has; the seam's JSDoc pins the single-decision-slot convention instead of inventing a provider registry.
|
||||
- **An inline `tools/pre-execute` permission gate in the ACP bridge** — rejected: prompting for every bridge-owned call hardwires the asking policy into the transport, cannot serve a second asker (sandbox escalation happens after execution starts, with no pre-execute moment), and leaves hook-produced `ask` decisions without a shared mechanism.
|
||||
- **The generic user-interaction seam (`ctx.userInteraction`)** — rejected as the approval mechanism: the two share a skeleton (route by agent, block for a human, handle absence), but approval's contract is narrower in every dimension that matters: a closed outcome vocabulary instead of free text, a protocol-native prompt attached to a tool call instead of a generic form, mandatory fail-closed absence, and audit events. Approval therefore does not ride the shipped `packages/interaction/user-interaction` / `ask_user_question` elicitation path — an elicitation form is not a permission prompt, and a free-text answer is not a closed outcome; sharing provider plumbing stays open if the two ever converge.
|
||||
- **Static optional injection in `dsh-tools`** — rejected: the vendored cordis `Inject` type has no optional flag — the object form maps service names to intercept config, and a declared inject gates the fiber. `ctx.get('approval')` is the documented opportunistic-consumption pattern (the `tool-bash` owner-token lookup, the loop's persistence probe), reads presence per call, and degrades correctly across HMR without extra machinery.
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-06-explicit-tool-order.md
|
||||
2026-07-06-explicit-tool-order.md: 7a4553b50da409b531a9fe14fcfbb201a5165fc4
|
||||
2026-07-06-explicit-tool-order.md: 4883a94d4271377dbfc7b6479d87d38ee9f710ed
|
||||
2026-07-06-explicit-tool-order.zh.md: a78a38a6ada3ed0a2f3a756830747916446e188c
|
||||
|
||||
@@ -31,7 +31,7 @@ Config plumbing follows the `persona` precedent, and `toolOrder` sits beside it:
|
||||
- **A linearization of the plugin dependency graph** — the relation is partial and independent tool plugins are incomparable; the flake happened with the partial order fully satisfied.
|
||||
- **Per-plugin `weight` on each tool contribution** — scatters the order across plugins yet still needs a global numbering convention nobody owns (the section `order` bands show that coordination cost being paid by hand).
|
||||
- **Sorting in `ToolRegistry.schemas()` (the registry layer)** — equally deterministic, but the registry is a membership store consumed by more than the assembly; ordering is a prompt-composition concern, and the assembly already owns the composition policy for sections.
|
||||
- **A `LlmService` config + `orderTools()` method the loop calls before logging the header** — works, but adds a public service method and a loop edit solely to apply a policy at a distance; every future request composer must remember the call. Canonicalizing where the list is born makes an unordered list unrepresentable, with zero new surface.
|
||||
- **A `LlmService` config + `orderTools()` method the loop calls before logging the header** — works, but adds a public service method and a loop edit solely to apply a policy at a distance; every future request composer must remember the call. Canonicalizing where the list is born makes an unordered list unrepresentable, with zero new API.
|
||||
- **Normalizing inside `llm.stream()`** — runs after the header event is logged (the flake survives) and rebuilds the deep-frozen envelope, silently disarming the reconstruction invariant.
|
||||
- **An exhaustive list (no rest entry)** — every newly loaded tool plugin would break boot; the mandatory rest entry keeps unlisted tools deterministic and their position explicit.
|
||||
- **A boot-time validation pass (a `SystemPrompt.assertToolOrderSatisfied()` called by `dsh-app-boot` after `loader.await()`)** — would turn the misconfiguration into a startup death instead of a first-turn failure, but costs a public service method plus a structural coupling from the generic boot glue to one service, and cannot replace the assembly-time check anyway (embedded callers never run app boot; registrations change after boot). No existing event can host the check either: cordis v4 has no ready-like event, `loader/entry-init`/`internal/status` fire mid-load (racy against tool registration, the very entropy this Agent Note kills), and the agent lifecycle events are no earlier than the assembly. One enforcement point at `assemble()` was judged worth the later failure moment.
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-06-sandbox.md
|
||||
2026-07-06-sandbox.md: 0214a202a41983c76fa25e3a82e1cfaec70a0d55
|
||||
2026-07-06-sandbox.zh.md: bd8dcdb74f723a955ea2a1cc5b224ef2ded4a8d5
|
||||
2026-07-06-sandbox.md: d37e9efc5ce8b1267c74591cb9edc01bef68ab8d
|
||||
2026-07-06-sandbox.zh.md: d3b74d441f72fc4c218979c327546031fd1b24f1
|
||||
|
||||
@@ -146,7 +146,7 @@ Each phase gets its full design when picked up, validated against the code at th
|
||||
- **Auto-retry inside the same tool call** — rejected: a hidden re-entry the log cannot reconstruct: one `tool/call` would have produced two executions with different policies — the retry is a NEW logged call with its own arguments and result facts.
|
||||
- **Advertise the escalation fields unconditionally** — rejected: under `dsh-bash-local` they are a dead lever — advertising an option the harness cannot honor manufactures doomed grants; capability-gating costs one registration-time read.
|
||||
- **A default-relative escalation ladder (advertise only the modes wider than the executor's registration-time default)** — rejected: per-session overrides make the default the wrong baseline — a session switched narrower than the default loses exactly the lever it needs, and under a `danger-full-access` default the fields vanish entirely while a `read-only`-overridden session stays confined with no escalation path. The enum pins the closed target vocabulary; strict widening is a per-call execution check against the session's effective mode.
|
||||
- **Per-session dynamic tool schemas** — rejected: schemas are registry-global by design (one assembly vocabulary, the pinned-header snapshot contract), and re-registering per session would buy only what the execution-time strict-wider check already guarantees, at the cost of a per-session schema surface and header churn on every switch.
|
||||
- **Per-session dynamic tool schemas** — rejected: schemas are registry-global by design (one assembly vocabulary, the pinned-header snapshot contract), and re-registering per session would buy only what the execution-time strict-wider check already guarantees, at the cost of a per-session schema and header churn on every switch.
|
||||
- **Hard-match the retry to a prior denial** — rejected: command-string identity is fragile (quoting, `workdir`, env prefixes, a pipeline retried as its failing stage) — false-rejects honest retries or is trivially satisfied; the real boundary is the human seeing command + justification. Revisit only if `allow_always` grant storage ever needs machine-checkable scopes.
|
||||
- **A generic `env/state` facts map with an owner service** — rejected: approval and sandbox compose independently, so neither's state may drag in a third package; single-key folds are one `findLast` each, dissolving the owner service; no invariant spans the knobs, so atomic multi-key patches bought nothing.
|
||||
- **Narrate each switch through `agent.inject()` plus a bus event** — rejected: independent notices expose owner ordering and intermediate combinations, while one pre-step composition can enqueue the complete current state atomically.
|
||||
@@ -171,7 +171,7 @@ Costs and accepted limits:
|
||||
|
||||
- **The one-wrapper illusion is given up knowingly.** A `tools/pre-execute` wrapper plus prompt conventions does not solve sandbox approval — the correct design costs structured denials, native runner probes, per-call policy carriage, and consistent cross-family enforcement, and this design pays it.
|
||||
- **`read-only` became a cross-family boundary through a follow-up.** This RFC shipped bash-only enforcement; the [cross-family fs sandbox RFC](2026-07-14-cross-family-fs-sandbox.md) extends the same mode vocabulary to the filesystem tools through a sandboxed `ctx.fs` provider and relocates the mode/root config and the `sandbox/mode` override to `ctx.sandboxPolicy` (§ In-process tools).
|
||||
- **Windows has no backend.** Its chain slot is reserved empty — fail-closed, never a fallthrough; filling it is a deferred phase.
|
||||
- **Windows is a partial backend.** This RFC originally reserved an empty, fail-closed win32 chain; the later [Windows ACL sandbox decision](2026-08-08-windows-acl-restricted-token-sandbox.md) filled it with the restricted-token runner. Its Everyone and hard-link gaps are reported as `enforcement: 'partial'`, never promoted to the full promise.
|
||||
- **The Seatbelt rung leans on Apple's deprecated-but-shipped `sandbox-exec` CLI.** As darwin's sole candidate it is selected without probing, so a future removal under a usable workdir surfaces as a runner-attributable spawn failure and an executable refusal through its fatal signature — both become `SANDBOX_UNAVAILABLE`, and the command never runs; fail closed, never open.
|
||||
- **Landlock confinement is only as complete as the running kernel's ABI.** Reported as `enforcement: 'partial'` rather than refused — the deliberate trade that keeps the fallback available on older-kernel hosts.
|
||||
- **Runner attribution uses an in-band protocol.** Exit status plus stderr cannot cryptographically identify the writer, so a confined child can mimic a fatal runner line and status to cause an availability/diagnostic false attribution. The conjunction and exact notice exclusion reduce accidental matches; this is not a sandbox bypass because the child is already confined.
|
||||
|
||||
@@ -171,7 +171,7 @@ fs/web/todo 在进程内执行，因此它们的沙箱语义是各自能力边
|
||||
|
||||
- **单一包装的幻觉被有意放弃。**`tools/pre-execute` 包装加提示词约定无法解决沙箱批准——正确的设计需要结构化拒绝、原生 runner 探测、按调用策略承载和一致的跨工具族强制，本设计为此付出了代价。
|
||||
- **`read-only` 通过后续设计成为跨工具族边界。** 本 Agent Note 最初只交付 bash 强制；[跨工具族 fs 沙箱 Agent Note](2026-07-14-cross-family-fs-sandbox.md) 通过沙箱化的 `ctx.fs` 提供方把同一模式词汇扩展到文件系统工具，并将 mode/root 配置和 `sandbox/mode` 覆盖迁移到 `ctx.sandboxPolicy`（§ 进程内工具）。
|
||||
- **Windows 没有后端。** 其链槽保留为空——失败关闭，绝不穿透；填充它是延迟阶段。
|
||||
- **Windows 后端只提供部分强制执行。** 本 RFC 最初预留了一条空的、失败关闭的 win32 链；后续的 [Windows ACL 沙箱决策](2026-08-08-windows-acl-restricted-token-sandbox.md)以受限令牌 runner 填充了它。其 Everyone 与硬链接缺口报告为 `enforcement: 'partial'`，绝不提升为完整承诺。
|
||||
- **Seatbelt 层级依赖 Apple 已弃用但仍交付的 `sandbox-exec` CLI。** 作为 darwin 的唯一候选，它无需探测即被选中，因此在 workdir 可用时，未来移除会表现为可归因于 runner 的 spawn 失败，可执行文件拒绝则通过其致命签名体现——两者都会变为 `SANDBOX_UNAVAILABLE`，且命令绝不会运行；失败关闭，绝不开放。
|
||||
- **Landlock 约束的完整度取决于运行内核的 ABI。** 报告为 `enforcement: 'partial'` 而非拒绝——这是有意的权衡，使备选在旧内核主机上仍可用。
|
||||
- **Runner 归因使用带内协议。** 退出状态与 stderr 无法以密码学方式识别写入者，因此受限子进程可以模仿 runner 的致命诊断行和状态，造成可用性或诊断误归因。多项证据的合取与精确通知排除减少了意外匹配；这不是沙箱绕过，因为子进程已经受到限制。
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-07-mcp-client-plugin.md
|
||||
2026-07-07-mcp-client-plugin.md: 077d978d8815d759574f89ed524ab3bc8c24a267
|
||||
2026-07-07-mcp-client-plugin.md: 24828586645778294ad7bdafc6fd13a9bfb6745f
|
||||
2026-07-07-mcp-client-plugin.zh.md: 8f58c9359ca8447717cc353f97f1333370fa6fda
|
||||
|
||||
@@ -200,7 +200,7 @@ Coverage is named per tier; each behavior lives at the cheapest tier that can ex
|
||||
## Consequences
|
||||
|
||||
- A `cordis.yml` entry per MCP server is the entire integration cost: `serverName: filesystem` + a stdio command (or a Streamable HTTP URL) puts `mcp__filesystem__read_file` in the model's tool list, callable, with the raw `read_file` on the wire.
|
||||
- Public names are part of session history and permission/config surfaces; the naming algorithm is a v1 contract pinned by tests, and changing it after release is a breaking change.
|
||||
- Public names are part of session history and permission/configuration APIs; the naming algorithm is a v1 contract pinned by tests, and changing it after release is a breaking change.
|
||||
- The `mcp__<serverName>__` qualifier costs tokens on every name. Accepted: descriptions and JSON schemas dominate tool-definition tokens, and the qualifier buys stable identity, collision isolation, and MCP-wide policy shapes (`mcp__*`, `mcp__github__*`).
|
||||
- **MCP SDK stability**: the `@modelcontextprotocol/sdk` is still evolving; breaking changes require updating the bridge. The version is pinned, and the SDK is widely adopted (Claude Desktop, Cursor, VS Code) so breaking changes are unlikely to be silent.
|
||||
- **Tool schema quality**: MCP servers may expose poorly-described tools (vague descriptions, incomplete JSON schemas). The harness passes them through as-is — garbage-in-garbage-out; that is the server author's responsibility, not the bridge's.
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-08-background-subagent-tasks.md
|
||||
2026-07-08-background-subagent-tasks.md: 679c57c3173a32b8cfccc6611b4ba74487e7f7e3
|
||||
2026-07-08-background-subagent-tasks.zh.md: c46f3925bc3334ff4cee79e4d9c95589e5361eaf
|
||||
2026-07-08-background-subagent-tasks.md: 2c2080d3520f8f784102af1344bea09875e6e61c
|
||||
2026-07-08-background-subagent-tasks.zh.md: 64c8bf898532ab8ab754dc950660df69fe2c39ad
|
||||
|
||||
@@ -18,7 +18,7 @@ Background subagents use the [generic background task runtime](../architecture/2
|
||||
|
||||
Foreground calls retain their synchronous contract: await provider startup and `run.result`, return final text only for `completed`, map other terminal reasons to an errored tool result, and always dispose the run before returning.
|
||||
|
||||
For a background call, the tool validates the parent and refuses an already-aborted execution signal before calling `ctx.tasks.start()`. The task runtime preflights the control surface and owner cleanup before invoking the producer starter. That starter creates an independent `AbortController` and begins `ctx.subagents.start()`; after the id is returned, the tool-call signal no longer owns the child.
|
||||
For a background call, the tool validates the parent and refuses an already-aborted execution signal before calling `ctx.tasks.start()`. The task runtime preflights the control API and owner cleanup before invoking the producer starter. That starter creates an independent `AbortController` and begins `ctx.subagents.start()`; after the id is returned, the tool-call signal no longer owns the child.
|
||||
|
||||
The task registration maps the subagent seam as follows:
|
||||
|
||||
@@ -53,11 +53,11 @@ Agents and logs may be session-scoped, but the task registry and predictable ids
|
||||
|
||||
### Incremental child transcript output
|
||||
|
||||
Streaming child history into the parent would blur the log boundary and make provider behavior diverge. This surface exposes final output only; richer observation belongs to session or UI tooling.
|
||||
Streaming child history into the parent would blur the log boundary and make provider behavior diverge. This tool exposes final output only; richer observation belongs to session or UI tooling.
|
||||
|
||||
## Testing
|
||||
|
||||
Unit coverage pins stop-reason mapping, dispose-before-report behavior, startup and result failures, pre-aborted refusal, detachment from the starting call's signal, cancellation before and after provider publication, collection through the real task tools, the no-surface preflight fence, missing-runtime failure, and per-instance schema gating. Snapshot coverage pins the model-facing schemas.
|
||||
Unit coverage pins stop-reason mapping, dispose-before-report behavior, startup and result failures, pre-aborted refusal, detachment from the starting call's signal, cancellation before and after provider publication, collection through the real task tools, the no-controller preflight fence, missing-runtime failure, and per-instance schema gating. Snapshot coverage pins the model-facing schemas.
|
||||
|
||||
## Consequences
|
||||
|
||||
|
||||
@@ -18,7 +18,7 @@ subagent 需要与其他长时间运行的工具相同的启动、收集、列
|
||||
|
||||
前台调用保留其同步约定：等待提供方启动和 `run.result`；仅当状态为 `completed` 时返回最终文本；将其他终止原因映射为出错的工具结果；并且始终在返回前释放该运行。
|
||||
|
||||
对于后台调用，工具会验证父级，并在调用 `ctx.tasks.start()` 前拒绝已中止的执行信号。任务运行时会在调用生产者启动器前，预检控制接口和拥有者清理。该启动器创建独立的 `AbortController` 并启动 `ctx.subagents.start()`；返回 id 之后，工具调用的信号不再拥有该子级。
|
||||
对于后台调用，工具会验证父级，并在调用 `ctx.tasks.start()` 前拒绝已中止的执行信号。任务运行时会在调用生产者启动器前，预检控制器和拥有者清理。该启动器创建独立的 `AbortController` 并启动 `ctx.subagents.start()`；返回 id 之后，工具调用的信号不再拥有该子级。
|
||||
|
||||
任务注册按以下方式映射 subagent seam：
|
||||
|
||||
@@ -57,7 +57,7 @@ agent 和日志可能以会话为作用域，但任务注册表和可预测 id
|
||||
|
||||
## 测试
|
||||
|
||||
单元测试覆盖固定了停止原因映射、在报告前释放资源、启动与结果失败、对预中止的拒绝、从启动调用信号分离、在提供方发布前后取消、通过真实任务工具收集、无控制接口的预检防线、运行时缺失失败，以及每实例 schema 开关。快照覆盖固定了面向模型的 schema。
|
||||
单元测试覆盖固定了停止原因映射、在报告前释放资源、启动与结果失败、对预中止的拒绝、从启动调用信号分离、在提供方发布前后取消、通过真实任务工具收集、无控制器的预检防线、运行时缺失失败，以及每实例 schema 开关。快照覆盖固定了面向模型的 schema。
|
||||
|
||||
## 影响
|
||||
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-08-self-referential-cordis-toolset.md
|
||||
2026-07-08-self-referential-cordis-toolset.md: 0d78e0adff487edae00c2422acc1ef8941e7636a
|
||||
2026-07-08-self-referential-cordis-toolset.md: 80f9a402562940727760a3686eb3c24866a053af
|
||||
2026-07-08-self-referential-cordis-toolset.zh.md: 665387a9fb24bf0e6fc04fdfb1ced88b932742ad
|
||||
|
||||
@@ -28,7 +28,7 @@ The vm isolates accidental global pollution, and the context façade hides frame
|
||||
|
||||
### Sandbox semantics
|
||||
|
||||
Mount code runs as an async-function body in a fresh vm realm. Its documented surface steers file, network, process, and timer access through Cordis services so mounts remain inspectable and disposable. Host-realm helpers still make Node escape possible, consistent with the trusted posture. `vmTimeoutMs` bounds only synchronous evaluation.
|
||||
Mount code runs as an async-function body in a fresh vm realm. Its documented API steers file, network, process, and timer access through Cordis services so mounts remain inspectable and disposable. Host-realm helpers still make Node escape possible, consistent with the trusted posture. `vmTimeoutMs` bounds only synchronous evaluation.
|
||||
|
||||
Sandbox globals are deliberately small: a tagged write-through `console` (`[cordis:<id>] …` on the host stdout/stderr, so a listener that fires long after the mount call still lands somewhere the user sees), the `harness.defineTool` / `harness.registerTool` registration pair, the encoding primitives fresh vm contexts lack (`btoa`/`atob` as host closures over `Buffer` — a sanctioned exception, `Buffer` itself is never exposed — plus `TextEncoder`/`TextDecoder`), and callable traps over the withheld Node APIs (`require`, `setTimeout`/`setInterval`/`setImmediate`/`clearTimeout`/`clearInterval`, `fetch`) that throw a redirect naming the cordis alternative. Only function-shaped globals are trapped; `process` and `Buffer` stay `undefined` so a `typeof` feature probe stays inert rather than detonating a throwing accessor.
|
||||
|
||||
@@ -48,7 +48,7 @@ Mounts relate to each other through ordinary cordis service semantics, with thei
|
||||
|
||||
### The generated API catalog
|
||||
|
||||
`cordis_inspect` serves API and event data from a generated catalog rather than a duplicated table. The generator reuses the Cordis catalog AST scan and emits service summaries, signatures, original service-method and event JSDoc, event modes, referenced type declarations, and the inherited context surface. Ambiguous type names are omitted and oversized declarations are marked as truncated.
|
||||
`cordis_inspect` serves API and event data from a generated catalog rather than a duplicated table. The generator reuses the Cordis catalog AST scan and emits service summaries, signatures, original service-method and event JSDoc, event modes, referenced type declarations, and the inherited context API. Ambiguous type names are omitted and oversized declarations are marked as truncated.
|
||||
|
||||
Freshness is gated like every generated artifact: `pnpm run verify-cordis-api` (in `doc-sync`) regenerates in memory and fails on any diff, so a JSDoc or public-signature edit cannot ship without regenerating the catalog the model reads. At runtime the inspect tool intersects the catalog with the live runtime rather than dumping it: broad reports render live catalogued services as summary + signatures, live services without a catalog entry (mount-provided ones) as name + owning fiber, catalogued services with no live provider tersely, and then the referenced type shapes. Exact-name reports render one live service or event with the original JSDoc immediately before each signature; keeping that detail opt-in avoids charging its token cost on exploratory listings.
|
||||
|
||||
@@ -66,7 +66,7 @@ Model-visible ⟺ logged holds with no new session event type: mount and unmount
|
||||
|---|---|---|
|
||||
| Schema correctness | `parameters` is still model-written JSON needing unified-schema validation, merely one step earlier | The same validation runs at the sandbox boundary, with the same instructive errors |
|
||||
| The code field | An `execute` body is still model-written JS in a vm; the realm and service-call correctness problems are unchanged | One sandbox, one normalization path, one guarded registration |
|
||||
| Capability coverage | Tools only; listeners, services, `inject` relations each need another structured tool — a surface that grows without bound | One vocabulary (a cordis plugin) covers every effect, present and future |
|
||||
| Capability coverage | Tools only; listeners, services, `inject` relations each need another structured tool — an API that grows without bound | One vocabulary (a cordis plugin) covers every effect, present and future |
|
||||
| Cross-mount composition | Not expressible in a tool-registration payload | Native `provide`/`inject`, ordinary cordis semantics |
|
||||
| Inspectability | Registers something the plugin list cannot show as a plugin | What the model mounts is exactly what `cordis_inspect` renders |
|
||||
| Model ergonomics | Wins for the single most common case (no plugin boilerplate) | Mitigated by the canonical recipe in the mount description plus boundary errors that teach the fix |
|
||||
@@ -77,7 +77,7 @@ The correctness investment therefore goes where it pays for every capability at
|
||||
|
||||
**A new `cordis/mount` session event.** A durable event recording each mount's source and name has clear precedent (`hook/invoked`, `compact/start`). It was declined for v1: mount and unmount are already visible as `tool/call` / `tool/result` pairs and the tool-set change is already logged as a full changed request header, so a dedicated event would only duplicate the record. It remains addable if an audit use case needs the mount source and name outside the tool call.
|
||||
|
||||
**A hardened / capability-restricted sandbox.** Trapping Node built-ins and handing mount code a whitelist façade rather than the raw context might suggest an intent to sandbox for safety. It is explicitly not that: the traps and the façade narrow the *surface* mount code sees — steering it onto cordis services and away from leak-prone Node built-ins and framework internals — for correctness and to close the unguarded-context escape, but the capabilities the façade exposes (`ctx.bash`, `ctx.fs`, `ctx.web`) reach the real runtime, so it is not a security boundary. A real one (separate process, permission prompts) was out of scope for a dev/opt-in toolset and would fight the entire point — handing the model the live runtime.
|
||||
**A hardened / capability-restricted sandbox.** Trapping Node built-ins and handing mount code a whitelist façade rather than the raw context might suggest an intent to sandbox for safety. It is explicitly not that: the traps and the façade narrow the *API* mount code sees — steering it onto cordis services and away from leak-prone Node built-ins and framework internals — for correctness and to close the unguarded-context escape, but the capabilities the façade exposes (`ctx.bash`, `ctx.fs`, `ctx.web`) reach the real runtime, so it is not a security boundary. A real one (separate process, permission prompts) was out of scope for a dev/opt-in toolset and would fight the entire point — handing the model the live runtime.
|
||||
|
||||
## Consequences
|
||||
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-10-agent-session-identity-and-log-location.md
|
||||
2026-07-10-agent-session-identity-and-log-location.md: 7b51ae41ac00c12a496940c891092580003646fa
|
||||
2026-07-10-agent-session-identity-and-log-location.md: 0b21193dbbfb05f78eb845f4a6893f6011584497
|
||||
2026-07-10-agent-session-identity-and-log-location.zh.md: b310db258f781cb5326e4bfe9e9daf3a6c1793e7
|
||||
|
||||
@@ -29,7 +29,7 @@ interface SessionPersistence {
|
||||
|
||||
`path` is an absolute local path to the backend's dedicated log for `meta`; `kind` identifies the representation. JSONL returns `{ kind: 'jsonl', path }` using its resolved root and path helpers. SQLite and any backend without an honest local per-session artifact return `undefined`. The query creates and flushes nothing, so it can report a lazy target path before that file exists.
|
||||
|
||||
The model-facing bash package owns a `ctx.bashEnv` registry. A contributor declares its stable name, every `DSH_*` key it may return, a description for each key, and `resolve(execution: ToolExecution)`. Duplicate contributor names, duplicate key ownership, reserved keys, malformed declarations, undeclared runtime output, and non-string output fail loudly. Registration is a Cordis effect and is removed with the contributing plugin fiber. `list()` exposes declarations without running resolvers, keeping the environment surface enumerable for diagnostics and future prompt/UI consumers.
|
||||
The model-facing bash package owns a `ctx.bashEnv` registry. A contributor declares its stable name, every `DSH_*` key it may return, a description for each key, and `resolve(execution: ToolExecution)`. Duplicate contributor names, duplicate key ownership, reserved keys, malformed declarations, undeclared runtime output, and non-string output fail loudly. Registration is a Cordis effect and is removed with the contributing plugin fiber. `list()` exposes declarations without running resolvers, keeping the environment API enumerable for diagnostics and future prompt/UI consumers.
|
||||
|
||||
The registry rebuilds a trusted overlay for every foreground and background bash `ToolExecution`:
|
||||
|
||||
@@ -48,7 +48,7 @@ The [Claude Code and Codex hook bridges](2026-06-30-hook-bridges.md) resolve tra
|
||||
|
||||
## Peer product findings
|
||||
|
||||
Peer products separate stable identity from physical storage. Codex injects stable `CODEX_THREAD_ID` into spawned shells while recorder and hook surfaces own transcript paths. Claude Code supplies `session_id` and `transcript_path` as structured hook/status input. OpenCode carries identity in structured tool context; Kimi Code expands a session placeholder; Reasonix keeps the active session path on its controller. The portable rule is to inject identity at the invocation boundary, let storage resolve location, and never use a process-global current-session variable in a concurrent harness.
|
||||
Peer products separate stable identity from physical storage. Codex injects stable `CODEX_THREAD_ID` into spawned shells while recorder and hook integrations own transcript paths. Claude Code supplies `session_id` and `transcript_path` as structured hook/status input. OpenCode carries identity in structured tool context; Kimi Code expands a session placeholder; Reasonix keeps the active session path on its controller. The portable rule is to inject identity at the invocation boundary, let storage resolve location, and never use a process-global current-session variable in a concurrent harness.
|
||||
|
||||
## Lifecycle and persistence semantics
|
||||
|
||||
@@ -78,7 +78,7 @@ A keyless full-loop integration drives the real agent loop, JSONL persistence, t
|
||||
|
||||
**Have each persistence backend register bash env directly.** That reverses the dependency from storage into one consumer and forces bash into deployments that do not use it. `locate()` is also still required by hooks.
|
||||
|
||||
**A model-facing `session_info` tool.** It adds schema and another call while bash already supplies the query surface; the registry generalizes to future environment facts without one tool per fact.
|
||||
**A model-facing `session_info` tool.** It adds schema and another call while bash already supplies the query API; the registry generalizes to future environment facts without one tool per fact.
|
||||
|
||||
## Consequences
|
||||
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-16-harness-level-loop.md
|
||||
2026-07-16-harness-level-loop.md: e5dd94fbf76e27f0843666f29622969635ef2fc3
|
||||
2026-07-16-harness-level-loop.zh.md: 773b5ff1f4b07206d0d817c29774435ce24407b4
|
||||
2026-07-16-harness-level-loop.md: 8fb89c00a18c77e8f1d14eb116fd3bd7d5789fad
|
||||
2026-07-16-harness-level-loop.zh.md: 88a455f22942e406ee0e4168c39131d4469c1b07
|
||||
|
||||
@@ -64,7 +64,7 @@ Only an admitted positive-round goal-sourced `user/message` charges a round. A s
|
||||
|
||||
Normal turn completion schedules another round only while the goal remains active, armed, and below its cap. Cancellation pauses. Rate limiting or quota exhaustion blocks with code `usage-limited`; cap exhaustion blocks with `round-limit`; queue failure uses `queue-failed`; turn errors, max-token stops, policy rejection, and unknown terminal results use their corresponding blocker codes. An independently composed request-recovery plugin may retry transient provider failures within that same turn; the goal driver never invents another round after an abnormal terminal outcome. A human can later authorize resume through ordinary language or `/goal resume`.
|
||||
|
||||
### Human and model surfaces
|
||||
### Human and model interactions
|
||||
|
||||
The human UX follows the compact Codex shape in the [public OpenAI Codex TUI dispatcher at commit `678157a`](https://github.com/openai/codex/blob/678157acaa819d5510adfe359abb5d0392cfe461/codex-rs/tui/src/chatwidget/slash_dispatch.rs#L750-L805): `/goal` shows status, `/goal <objective>` creates, and `edit`, `pause`, `resume`, or `clear` perform direct lifecycle actions. The commit permalink keeps the researched grammar verifiable as Codex evolves. Status includes durable phase, admitted/capped rounds, and live armed/disarmed activation. Direct status and command output do not enter model history; accepted domain mutations remain reconstructable because the goal service records them.
|
||||
|
||||
@@ -114,7 +114,7 @@ The six owning Agent Notes record unit, integration, process, snapshot, cancella
|
||||
- Humans receive a small Codex-shaped UX; models receive a compact tool set whose mutating calls require a direct human message in the current live root-agent turn; deployments can remove either independently.
|
||||
- Ralph demonstrates a nontrivial fixed policy entirely as a plugin over existing workflow and subagent primitives.
|
||||
- Round limits are generous by default but remain deployment-controlled. They bound iterations, not tokens, price, elapsed time, or external side effects.
|
||||
- The original proposal's evaluator, budget, reflector, background-task, CLI, and generic loop-session architecture is intentionally not part of the implemented public surface.
|
||||
- The original proposal's evaluator, budget, reflector, background-task, CLI, and generic loop-session architecture is intentionally not part of the implemented public API.
|
||||
|
||||
## Known limitations and deferred work
|
||||
|
||||
@@ -124,6 +124,6 @@ The six owning Agent Notes record unit, integration, process, snapshot, cancella
|
||||
- **No time scheduler** — interval `/loop`, cron, proactive maintenance, and cloud or desktop scheduling are outside this decision.
|
||||
- **No generic loop journal or execution-world rewind** — session replay reconstructs goal history, not prior files, processes, environment, credentials, or external side effects. Ralph treats the current workspace as authority and carries no cross-run journal.
|
||||
- **No goal reflector** — concern events, automatic no-progress heuristics, goal revision by an independent reflector, stuck-pattern detection, and `loop_split` are not implemented. Humans can edit, pause, clear, or resume the goal directly.
|
||||
- **Ralph policy remains narrow** — one round creates one fresh child; within-round fan-out, evaluator/worker role separation, dynamic provider/model selection, and structural recursive-Ralph tool denial need separate policy surfaces. Prompt guidance is not enforcement.
|
||||
- **Ralph policy remains narrow** — one round creates one fresh child; within-round fan-out, evaluator/worker role separation, dynamic provider/model selection, and structural recursive-Ralph tool denial need separate policy APIs. Prompt guidance is not enforcement.
|
||||
- **Ralph does not retry a failed child** — an ordinary failure preserves the failed round and last good handoff, while fatal workflow infrastructure failures can end before that state is available. Retry count, backoff, and richer failure transport need separate policy and boundary design.
|
||||
- **Portable UI remains modest** — TUI renders plain-text goal status and generic Ralph cards. ACP carries only committed assistant text; there is no continuous status widget, reconnectable command output, modal goal editor, or command plane in ACP, the headless CLI, or JSON-RPC.
|
||||
|
||||
@@ -64,7 +64,7 @@ Goal Round 驱动器为每个特定的实时 agent 至多拥有一个待定预
|
||||
|
||||
普通 Turn 完成后，只有目标仍活跃、已激活且低于上限时才会安排另一个 Round。取消会暂停。速率限制或配额耗尽以代码 `usage-limited` 阻塞；上限耗尽使用 `round-limit`；队列失败使用 `queue-failed`；Turn 错误、max-token 停止、策略拒绝与未知终止结果使用各自对应的阻塞代码。独立组合的请求恢复插件可以在同一个 Turn 内重试暂时性 provider 失败；目标驱动器绝不会在异常终止结果后凭空发起另一个 Round。人类随后可以通过普通语言或 `/goal resume` 授权恢复。
|
||||
|
||||
### 人类与模型表面
|
||||
### 人类与模型交互
|
||||
|
||||
人类 UX 遵循 [OpenAI Codex 在提交 `678157a` 时的公开 TUI 分发器](https://github.com/openai/codex/blob/678157acaa819d5510adfe359abb5d0392cfe461/codex-rs/tui/src/chatwidget/slash_dispatch.rs#L750-L805)中的紧凑形态：`/goal` 显示状态，`/goal <objective>` 创建目标，而 `edit`、`pause`、`resume` 或 `clear` 执行直接生命周期操作。该提交永久链接让研究所得语法在 Codex 演进时仍可验证。状态包含持久阶段、已接纳/上限 Round 数以及实时已激活/未激活状态。直接状态与命令输出不会进入模型历史；已接受领域变更仍可重建，因为目标服务会记录它们。
|
||||
|
||||
@@ -114,7 +114,7 @@ Codex 提供了这里采用的最小可观察目标 UX：一个附着于聊天
|
||||
- 人类获得小型 Codex 形态 UX；模型获得一组紧凑工具，其中的修改操作要求当前实时根 agent Turn 中有一条人类直接发送的消息；部署可以独立移除任一能力。
|
||||
- Ralph 展示了非平凡固定策略可以完全作为现有 workflow 与 subagent 原语之上的插件实现。
|
||||
- Round 上限默认宽裕，但仍由部署控制。它限制迭代次数，不限制 token、价格、耗时或外部副作用。
|
||||
- 原始提案中的评估器、预算、反思器、后台任务、CLI 与通用 loop-session 架构有意不进入已实现公开表面。
|
||||
- 原始提案中的评估器、预算、反思器、后台任务、CLI 与通用 loop-session 架构有意不进入已实现的公开 API。
|
||||
|
||||
## 已知限制与暂缓事项
|
||||
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-19-fresh-agent-ralph-workflow-tool.md
|
||||
2026-07-19-fresh-agent-ralph-workflow-tool.md: bbc8aba54bbd2cbfe9d9ba1fcc8cf9ad5c257d4f
|
||||
2026-07-19-fresh-agent-ralph-workflow-tool.zh.md: 7ac405c2f6f80fa3151fb8b88c266ae2b83ea3eb
|
||||
2026-07-19-fresh-agent-ralph-workflow-tool.md: e0f152d0e46eebe09d9d961eb241aa18cca403d0
|
||||
2026-07-19-fresh-agent-ralph-workflow-tool.zh.md: 02b1f0421263db928aff354d04abaf6bdced3249
|
||||
|
||||
@@ -36,7 +36,7 @@ The fixed prompt passes only the immutable objective, current round and cap, a w
|
||||
|
||||
The workflow language maps a normally settled but unsuccessful child to `null`. The fixed script detects that value before report validation and returns `round-failed` with the failed round plus the last successful handoff when one exists; the tool turns it into an error instead of misclassifying it as a malformed report or budget exhaustion. Ralph adds no retry policy. Fatal provider-start, transport, worker, and workflow errors remain generic workflow failures because the workflow seam does not carry a recoverable child report on those paths.
|
||||
|
||||
### Model and UI surface
|
||||
### Model and UI
|
||||
|
||||
The model may supply only `objective` and optional `maxRounds`; provider selection, report schema, handoff cap, and script are deployment-owned. A fixed prompt section says to use `ralph` only when the direct human explicitly asks for Ralph or fresh-agent iteration, and distinguishes it from same-session goals, bounded delegation, and general fan-out workflows. This is guidance rather than a new goal UX state machine.
|
||||
|
||||
@@ -46,12 +46,12 @@ Human-facing presentation uses a generic `ralph` card whose raw input is the obj
|
||||
|
||||
Unit tests cover config and call-cap resolution, provider capability rejection, fixed start-request routing and child ceiling, all successful terminal outcomes, ordinary child-failure envelopes, malformed and oversized boundary values, exact successful-result truncation, abort timing, disposal, render intent, prompt lifecycle, and namespace-plugin shape at per-file 100% coverage. Worker-engine tests prove synchronous provider-route validation, per-run child ceilings below the deployment ceiling, and that a provider override selects every child without changing the configured default, including the built `lib/worker.cjs` under plain Node.
|
||||
|
||||
A keyless real-stack integration drives the fixed script through the actual worker-thread engine, spawn provider, structured-output runtime, and agent loop. It proves distinct child identities, absent `seedLength`, inherited cwd, no parent-history markers in either child request, exact previous-report handoff only in the following round, one phase event, terminal completion, and disposal of both children. The same real stack covers blocker and round-limit outcomes, unnormalized and semantically invalid reports, oversized handoffs, ordinary child failure with the last good handoff, and cancellation to child quiescence. A shipped keyless headless snapshot additionally boots the real `examples/headless-agent` composition, invokes `ralph`, pins the parent stream transcript, and inspects persisted logs for two distinct unseeded child sessions and the round-one handoff appearing only in round two. Tool tests pin generic call/result presentation, while ACP replay header snapshots pin the shipped schema and prompt-guidance transcript surface.
|
||||
A keyless real-stack integration drives the fixed script through the actual worker-thread engine, spawn provider, structured-output runtime, and agent loop. It proves distinct child identities, absent `seedLength`, inherited cwd, no parent-history markers in either child request, exact previous-report handoff only in the following round, one phase event, terminal completion, and disposal of both children. The same real stack covers blocker and round-limit outcomes, unnormalized and semantically invalid reports, oversized handoffs, ordinary child failure with the last good handoff, and cancellation to child quiescence. A shipped keyless headless snapshot additionally boots the real `examples/headless-agent` composition, invokes `ralph`, pins the parent stream transcript, and inspects persisted logs for two distinct unseeded child sessions and the round-one handoff appearing only in round two. Tool tests pin generic call/result presentation, while ACP replay header snapshots pin the shipped schema and prompt-guidance transcript output.
|
||||
|
||||
## Alternatives considered
|
||||
|
||||
- **Put Ralph in the same-session goal driver** — rejected because goal rounds intentionally preserve one conversation, while Ralph's defining property is a fresh context per round; combining them would make goal lifecycle and child orchestration inseparable.
|
||||
- **Expose a `fresh` or loop flag on the general workflow tool** — rejected because the model-written script surface should remain general and provider-neutral; Ralph's fixed report protocol and stop policy deserve one reviewable consumer.
|
||||
- **Expose a `fresh` or loop flag on the general workflow tool** — rejected because the model-written script API should remain general and provider-neutral; Ralph's fixed report protocol and stop policy deserve one reviewable consumer.
|
||||
- **Use `subagent_fork` for replay convenience** — rejected because inherited completed turns are implicit, growing handoff state and violate the fresh-context contract. The workspace plus one structured report is replayable without inserting artificial cancellation records.
|
||||
- **Call the subagent seam directly from the tool** — rejected because the existing workflow engine already owns foreground orchestration, structured children, cancellation propagation, worker termination, events, and quiescent disposal. Reusing it demonstrates plugin composition instead of building a second loop runtime.
|
||||
- **Silently truncate a large report** — rejected because truncation can remove status evidence or next steps while still looking like an authoritative handoff. A producer must emit a valid report within the configured bound.
|
||||
@@ -62,7 +62,7 @@ A keyless real-stack integration drives the fixed script through the actual work
|
||||
- Goal rounds and Ralph rounds stay different concepts: the former is one same-session continuation turn, while the latter is one fresh child inside a foreground workflow.
|
||||
- The workspace becomes authoritative cross-round memory, so workers must inspect and verify it rather than trusting a narrative handoff.
|
||||
- A generous round ceiling permits substantial autonomous work, while deployment config still bounds child count and every handoff remains size-limited.
|
||||
- Provider routing and a lowerable per-run child ceiling become explicit workflow start concerns without expanding the script or ordinary workflow tool surface.
|
||||
- Provider routing and a lowerable per-run child ceiling become explicit workflow start concerns without expanding the script or ordinary workflow tool set.
|
||||
|
||||
## Known limitations and deferred work
|
||||
|
||||
@@ -71,4 +71,4 @@ A keyless real-stack integration drives the fixed script through the actual work
|
||||
- Round count is the only aggregate budget. Token, currency, elapsed-time, and provider-usage budgets remain separate future policy.
|
||||
- One round creates one child. Within-round fan-out, evaluator/worker role separation, dynamic provider or model selection, and cross-run journals are deferred.
|
||||
- An ordinary child failure ends the run without retry, while preserving the failed round and last successful handoff. Fatal workflow infrastructure failures can end before the fixed script returns that state; adding retry or richer failure transport requires separate policy and boundary design.
|
||||
- Prompt guidance asks models not to invoke Ralph recursively; a structural child-tool restriction would require a separately designed workflow child-policy surface.
|
||||
- Prompt guidance asks models not to invoke Ralph recursively; a structural child-tool restriction would require a separately designed workflow child-policy API.
|
||||
|
||||
@@ -36,7 +36,7 @@ Ralph 插件的 `subagentProvider` 默认为 `spawn`。每次调用前，它要
|
||||
|
||||
工作流语言会把正常结束但未成功的子 agent 映射为 `null`。固定脚本会在报告验证前检测该值，并返回 `round-failed`，其中包含失败的 Round，以及存在时的上一份成功交接；工具会把它转成错误，而不会误判为畸形报告或预算耗尽。Ralph 不添加重试策略。致命的提供方启动、传输、工作线程和工作流错误仍是通用工作流失败，因为这些路径上的 workflow seam 不携带可恢复的子报告。
|
||||
|
||||
### 模型与 UI 表面
|
||||
### 模型与 UI
|
||||
|
||||
模型只能提供 `objective` 和可选的 `maxRounds`；提供方选择、报告 schema、交接上限和脚本都由部署拥有。固定提示区段说明，只有直接交互的人类明确要求 Ralph 或全新 agent 迭代时才使用 `ralph`，并将其与同会话目标、有界委派和通用扇出工作流区分开。这是指导，而不是新的目标 UX 状态机。
|
||||
|
||||
@@ -71,4 +71,4 @@ Ralph 插件的 `subagentProvider` 默认为 `spawn`。每次调用前，它要
|
||||
- Round 数是唯一聚合预算。token、货币、耗时和提供方用量预算仍属于未来的独立策略。
|
||||
- 每个 Round 创建一个子 agent。Round 内扇出、evaluator/工作者角色分离、动态提供方或模型选择，以及跨运行日志均被推迟。
|
||||
- 普通子 agent 失败会结束运行且不重试，同时保留失败 Round 与上一份成功交接。致命工作流基础设施错误可能在固定脚本返回该状态前结束；增加重试或更丰富的失败传输需要独立的策略与边界设计。
|
||||
- 提示指导模型不要递归调用 Ralph；结构化的子 agent 工具限制需要另行设计工作流子策略表面。
|
||||
- 提示指导模型不要递归调用 Ralph；结构化的子 agent 工具限制需要另行设计工作流子策略 API。
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-19-human-goal-command.md
|
||||
2026-07-19-human-goal-command.md: d68e4025a4d37d07211f15ddfc6069bc7c637124
|
||||
2026-07-19-human-goal-command.md: 7ed104eef1a48879a6e93af707c6ea9d8f97190c
|
||||
2026-07-19-human-goal-command.zh.md: dbda35c731ccdc7aff2d4213a3df8b78070319df
|
||||
|
||||
@@ -32,7 +32,7 @@ Control words are ASCII-case-insensitive after outer whitespace trimming. They a
|
||||
|
||||
Status output omits branded ids and compare-and-set revisions because those are model/plugin coordination details rather than human controls. It includes activation because that fact changes whether work will continue, and a blocked goal includes its durable policy code and human-readable explanation. Command hints are derived from the exact state: an armed active goal offers pause, a disarmed active or paused/blocked goal offers resume, and a completed goal offers replacement or clear.
|
||||
|
||||
Expected `GoalError` failures become one stable, branded-id-free `CommandResult.error`, so domain diagnostics do not leak compare-and-set internals into the human surface and invalid operations never enter model history. The current status supplies the actionable state-specific recovery. Other exceptions remain adapter-visible command failures; treating programmer faults as ordinary domain errors would hide defects. The command handler performs only synchronous domain mutations, so request cancellation is decided by the command registry before the mutation begins and there is no escaped asynchronous side effect to unwind.
|
||||
Expected `GoalError` failures become one stable, branded-id-free `CommandResult.error`, so domain diagnostics do not leak compare-and-set internals into the UI and invalid operations never enter model history. The current status supplies the actionable state-specific recovery. Other exceptions remain adapter-visible command failures; treating programmer faults as ordinary domain errors would hide defects. The command handler performs only synchronous domain mutations, so request cancellation is decided by the command registry before the mutation begins and there is no escaped asynchronous side effect to unwind.
|
||||
|
||||
Generic slash input, status text, and errors are not persisted. Successful goal mutations append the domain-owned `goal/change` event and do not queue model context. The command introduces no second audit record that could disagree with the domain event.
|
||||
|
||||
@@ -50,7 +50,7 @@ The producer suite uses the real command registry, goal service, agent registry,
|
||||
|
||||
- **Let the model handle `/goal` as ordinary text** — rejected because status and direct lifecycle actions would cost a model turn, could be reinterpreted, and would not provide deterministic command discovery.
|
||||
- **Implement separate handlers in each UI** — rejected because grammar, error behavior, and goal-state formatting would drift and optional deployments could not add or remove the capability as one effect.
|
||||
- **Add modal editing and replacement confirmation to `ctx.commands`** — rejected because the existing cross-surface contract is unstructured input plus direct output; a general interaction protocol needs more than this one producer.
|
||||
- **Add modal editing and replacement confirmation to `ctx.commands`** — rejected because the existing cross-UI contract is unstructured input plus direct output; a general interaction protocol needs more than this one producer.
|
||||
- **Silently replace an unfinished goal** — rejected because it combines clear and create without atomicity or explicit destructive intent.
|
||||
- **Expose goal id and revision in human status** — rejected because human actions always target the exact current view inside one synchronous handler; those fields add implementation noise without preventing another race.
|
||||
- **Enable goals unconditionally in the UI-less spine** — rejected because one-shot SDK/CLI settlement is a physical-turn API, not a goal-operation API.
|
||||
@@ -65,7 +65,7 @@ The producer suite uses the real command registry, goal service, agent registry,
|
||||
|
||||
## Known limitations and deferred work
|
||||
|
||||
- The portable command contract has no modal editor or confirmation interaction; inline edit and explicit clear are intentional until a general cross-surface interaction primitive exists.
|
||||
- The portable command contract has no modal editor or confirmation interaction; inline edit and explicit clear are intentional until a general cross-UI interaction primitive exists.
|
||||
- `/goal` does not accept a per-command round cap. Deployment config owns the default, and the authorized model tool can edit a cap after direct human instruction.
|
||||
- TUI renders portable plain text rather than a continuously updated goal status widget. Reconnectable command output and adapter-specific status indicators are deferred.
|
||||
- The ACP automation server, headless CLI, and JSON-RPC entry points do not consume the command registry.
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-19-model-facing-goal-tools.md
|
||||
2026-07-19-model-facing-goal-tools.md: c2f1210ea636b61e99111d261cf9302c77576b46
|
||||
2026-07-19-model-facing-goal-tools.zh.md: fd8ca90ba3b6809a7225b118ec6d311fb193cc8f
|
||||
2026-07-19-model-facing-goal-tools.md: 2f2a1c2319dec96df9da83efbe92f1149001b8f2
|
||||
2026-07-19-model-facing-goal-tools.zh.md: 481ba1abb8ac66dc379d69f546e08138f42a0b22
|
||||
|
||||
@@ -6,9 +6,9 @@ English | [中文](2026-07-19-model-facing-goal-tools.zh.md)
|
||||
|
||||
## Problem
|
||||
|
||||
The persisted goal domain deliberately exposes lifecycle verbs to plugins, not directly to a model. A model still needs a small control surface for discovering the current goal, creating one from human intent, and changing its lifecycle. Prompt guidance alone cannot establish who authorized a mutation: a subagent, injected plugin message, stale model turn, or resumed session could all produce the same tool arguments.
|
||||
The persisted goal domain deliberately exposes lifecycle verbs to plugins, not directly to a model. A model still needs a small control API for discovering the current goal, creating one from human intent, and changing its lifecycle. Prompt guidance alone cannot establish who authorized a mutation: a subagent, injected plugin message, stale model turn, or resumed session could all produce the same tool arguments.
|
||||
|
||||
The surface also needs to preserve the separation between durable state and live execution authority. A restored or forked session can replay an active goal but starts disarmed; a later human request such as “continue” should let the model rearm it without requiring a literal command phrase. Conversely, an admitted autonomous goal round must be able to report completion or a persistent blocker without gaining permission to edit, pause, resume, or replace the human objective.
|
||||
The tool API also needs to preserve the separation between durable state and live execution authority. A restored or forked session can replay an active goal but starts disarmed; a later human request such as “continue” should let the model rearm it without requiring a literal command phrase. Conversely, an admitted autonomous goal round must be able to report completion or a persistent blocker without gaining permission to edit, pause, resume, or replace the human objective.
|
||||
|
||||
## Decision
|
||||
|
||||
@@ -43,7 +43,7 @@ Unit coverage pins registration and disposal, exclusive scheduling, generated pr
|
||||
## Alternatives considered
|
||||
|
||||
- **Rely on prompt instructions for authority** — rejected because text can guide model judgment but cannot authenticate the live caller, turn, or source event.
|
||||
- **Expose every goal-service verb as a separate tool** — rejected because a compact read/create/update surface reduces schema cost and keeps compare-and-set behavior uniform.
|
||||
- **Expose every goal-service verb as a separate tool** — rejected because a compact read/create/update API reduces schema cost and keeps compare-and-set behavior uniform.
|
||||
- **Require exact command phrases** — rejected because natural-language intent, including languages other than English, should be interpreted by the model; execution authority depends on a direct human message in the current turn rather than spelling.
|
||||
- **Authorize from persisted root or fork metadata** — rejected because a fork that becomes an independently resumed top-level session should accept new human authority, while a currently owned child should not.
|
||||
- **Let autonomous rounds edit or resume the goal** — rejected because continuation authority is narrower than authority to redefine or restart the human objective.
|
||||
@@ -52,7 +52,7 @@ Unit coverage pins registration and disposal, exclusive scheduling, generated pr
|
||||
|
||||
## Consequences
|
||||
|
||||
- Models receive a stable, compact lifecycle surface without direct access to the goal service.
|
||||
- Models receive a stable, compact lifecycle API without direct access to the goal service.
|
||||
- State-changing calls require a live runtime-root agent and a direct human message in the current turn, as well as durable compare-and-set references.
|
||||
- Human requests can create and rearm goals through ordinary natural language, while restored sessions remain inert until such input arrives.
|
||||
- Goal rounds can finish or report a repeated blocker but cannot broaden their own mandate.
|
||||
|
||||
@@ -6,7 +6,7 @@ Status: implemented
|
||||
|
||||
## 问题
|
||||
|
||||
持久目标领域有意把生命周期动词提供给插件，而不直接提供给模型。模型仍然需要一个小型控制面，用于发现当前目标、根据人类意图创建目标并改变其生命周期。仅靠提示词指导无法确定是谁授权了一次变更：子智能体、注入的插件消息、陈旧的模型轮次或恢复后的会话都可能产生相同的工具参数。
|
||||
持久目标领域有意把生命周期动词提供给插件，而不直接提供给模型。模型仍然需要一个小型控制 API，用于发现当前目标、根据人类意图创建目标并改变其生命周期。仅靠提示词指导无法确定是谁授权了一次变更：子智能体、注入的插件消息、陈旧的模型轮次或恢复后的会话都可能产生相同的工具参数。
|
||||
|
||||
该表面还需要保持持久状态与实时执行权限之间的分离。恢复或 fork（派生）后的会话可以回放活跃目标，但初始处于未激活状态；后续人类提出“继续”之类的请求时，模型应能重新激活目标，而无需用户使用字面命令。相反，已接纳的自主 Goal Round 必须能够报告完成或持续阻塞，却不能因此获得编辑、暂停、恢复或替换人类目标的权限。
|
||||
|
||||
@@ -43,7 +43,7 @@ Status: implemented
|
||||
## 考虑过的替代方案
|
||||
|
||||
- **依赖提示词指令实施权限**——不予采纳，因为文本可以指导模型判断，却不能认证实时调用者、轮次或来源事件。
|
||||
- **把每个目标服务动词分别暴露为工具**——不予采纳，因为紧凑的读取/创建/更新表面可以降低模式成本，并保持统一的比较并交换行为。
|
||||
- **把每个目标服务动词分别暴露为工具**——不予采纳，因为紧凑的读取/创建/更新 API 可以降低 schema 成本，并保持统一的比较并交换行为。
|
||||
- **要求精确命令短语**——不予采纳，因为自然语言意图（包括英语以外的语言）应由模型解释；执行权限取决于当前轮次中人类直接发送的消息，而不是拼写。
|
||||
- **根据持久的根或派生元数据授权**——不予采纳，因为成为独立恢复顶层会话的派生应接受新的人类权限，而当前仍受所有权约束的子智能体则不应接受。
|
||||
- **允许自主回合编辑或恢复目标**——不予采纳，因为继续执行权限比重新定义或重启人类目标的权限更窄。
|
||||
@@ -52,7 +52,7 @@ Status: implemented
|
||||
|
||||
## 后果
|
||||
|
||||
- 模型获得稳定而紧凑的生命周期表面，无需直接访问目标服务。
|
||||
- 模型获得稳定而紧凑的生命周期 API，无需直接访问目标服务。
|
||||
- 改变状态的调用要求实时运行时根 agent、当前轮次中人类直接发送的消息，以及持久比较并交换引用。
|
||||
- 人类可以通过普通自然语言请求创建和重新激活目标，而恢复后的会话在收到此类输入前保持静止。
|
||||
- Goal Round 可以完成或报告重复阻塞，但不能自行扩大任务权限。
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-19-persisted-same-session-goal-domain.md
|
||||
2026-07-19-persisted-same-session-goal-domain.md: cdbbdd55c8c9b73851d5e729492656eda1380430
|
||||
2026-07-19-persisted-same-session-goal-domain.md: bc6c9ec14c16b20bc8434bac0f1eb6070ea8920d
|
||||
2026-07-19-persisted-same-session-goal-domain.zh.md: 7c8af321ec78779ecbf4e5e00805ef6f152b2aa1
|
||||
|
||||
@@ -28,7 +28,7 @@ Incremental replay advances its cursor after each valid event and remains positi
|
||||
|
||||
At most one goal is current. Create requires no current non-complete goal and always generates a revision-one id not used earlier in the session; a completed goal may be replaced. Every other mutation carries the expected `GoalRef`, and stale ids or revisions reject. Resume accepts a paused or blocked phase, or a disarmed active goal, only when the round cap has remaining capacity. The domain validates blocker reason shape but deliberately leaves reason codes and the decision to block to policy consumers.
|
||||
|
||||
A cache built from any seed starts disarmed, and every `agent/session-start` edge disarms it again. `GoalService.disarm(agent)` also lets a lifecycle owner remove process-local authority without a session event, revision change, or `goal/changed` notification. Resume, fork, and continuation-driver replacement therefore preserve the durable objective and history but never initiate work on their own. A later human prompt can be interpreted by the model, whose policy surface may explicitly call resume and arm the goal.
|
||||
A cache built from any seed starts disarmed, and every `agent/session-start` edge disarms it again. `GoalService.disarm(agent)` also lets a lifecycle owner remove process-local authority without a session event, revision change, or `goal/changed` notification. Resume, fork, and continuation-driver replacement therefore preserve the durable objective and history but never initiate work on their own. A later human prompt can be interpreted by the model, whose policy API may explicitly call resume and arm the goal.
|
||||
|
||||
### Service boundary
|
||||
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-19-plugin-command-registration.md
|
||||
2026-07-19-plugin-command-registration.md: 76feba84492bd6b245246a6f1fb1e8f68d555684
|
||||
2026-07-19-plugin-command-registration.md: f2ea6fca14ef5a0d7057d2652ee5f511fc6c15e4
|
||||
2026-07-19-plugin-command-registration.zh.md: e89cb79ba1e15dfefa527fe7bc37c09c3449f5b4
|
||||
|
||||
@@ -24,7 +24,7 @@ A `CommandDefinition` contains a lowercase name without `/`, a non-empty descrip
|
||||
|
||||
### Scope and lifecycle
|
||||
|
||||
An unscoped registration is global. A command-injected plugin mounted beneath an agent context inherits that agent's scope key and lifetime, so its definition shadows a same-named global only for that exact agent. The child declares its own `commands` injection because `agent.ctx` intentionally inherits the core agent-loop dependency surface; adding a UI service to the loop merely to enable scoped registration would invert the dependency graph.
|
||||
An unscoped registration is global. A command-injected plugin mounted beneath an agent context inherits that agent's scope key and lifetime, so its definition shadows a same-named global only for that exact agent. The child declares its own `commands` injection because `agent.ctx` intentionally inherits the core agent-loop dependency API; adding a UI service to the loop merely to enable scoped registration would invert the dependency graph.
|
||||
|
||||
Registration and removal emit the unfiltered, non-vetoing `commands/change` registry notification. Adapters recompute each live agent's effective view rather than trying to infer which sessions a change affects. The registry contains and logs each observer failure independently, so a broken UI refresh cannot roll back another plugin's mutation or starve a later observer. Cordis ownership removes definitions when their producer, UI instance, or agent scope unloads, so HMR cannot leave stale discovery entries or handlers.
|
||||
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-20-code-mode-typed-tool-returns.md
|
||||
2026-07-20-code-mode-typed-tool-returns.md: 059c715e35fb856e92d35926e6133701e5602c97
|
||||
2026-07-20-code-mode-typed-tool-returns.md: 6b31dbca21a22a24f2bbb0ef61097622884165c3
|
||||
2026-07-20-code-mode-typed-tool-returns.zh.md: 1bb88d29ffd96d65ce06033202499d2780814abd
|
||||
|
||||
@@ -6,7 +6,7 @@ English | [中文](2026-07-20-code-mode-typed-tool-returns.zh.md)
|
||||
|
||||
## Problem
|
||||
|
||||
Code Mode originally projected each nested tool result back from `ContentBlock[]` into one string. That preserved the human-readable Native surface but erased the canonical result the tool had already produced: programs had to scrape task ids and dynamic mount ids from prose, structured search and workflow results lost their shape, and non-text blocks became placeholders. The generated SDK could describe arguments but could only promise `Promise<string>` regardless of the tool's real output.
|
||||
Code Mode originally projected each nested tool result back from `ContentBlock[]` into one string. That preserved the human-readable Native presentation but erased the canonical result the tool had already produced: programs had to scrape task ids and dynamic mount ids from prose, structured search and workflow results lost their shape, and non-text blocks became placeholders. The generated SDK could describe arguments but could only promise `Promise<string>` regardless of the tool's real output.
|
||||
|
||||
The runtime also treated binding values and the final program value as presentation data. Separate log and completion caps could replace an oversized or non-cloneable completion with inspected text even though intermediate values do not enter model context. That made programmatic composition lossy and confused the memory boundary with the prompt boundary.
|
||||
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-21-continuable-background-subagents.md
|
||||
2026-07-21-continuable-background-subagents.md: e37abdd798242bc1086754e86a85ef46c08e59a3
|
||||
2026-07-21-continuable-background-subagents.zh.md: 729ec62d9d3511661638144cdef6e31bf98ecdb7
|
||||
2026-07-21-continuable-background-subagents.md: 470b2328fd5c27e85ac2ca08b53f353f41b54fc9
|
||||
2026-07-21-continuable-background-subagents.zh.md: 7288fbdf916700f054451ce6a40eca2bd90f0866
|
||||
|
||||
@@ -37,7 +37,7 @@ Every later turn creates another Task. Its producer resources cover only that ac
|
||||
|
||||
Opening a child session in a human-facing adapter reads its persisted transcript and does not resume an Agent merely to display it. Human input starts or joins the same Task-backed activation used by parent input through the continuation manager. A human-started Task retains the exact currently loaded parent Agent as its notification target, and `task_output` remains the single result path. The existing completion listener injects at most one unsolicited notice while the Task is unreported; `kill`, a terminal read, or a terminal wait may mark it reported and suppress that notice. Human interaction is therefore permitted only while that parent instance remains live. A user-owned conversation that may outlive the parent and explicitly merge a conclusion back belongs to [interactive side sessions](../../proposed/feature/2026-07-08-interactive-side-sessions.md), not this Task-owned lifecycle.
|
||||
|
||||
`TaskService.start()` rejects producers when no Task control surface is attached. A human-facing adapter that accepts child input must therefore attach a Task control surface, or run in a deployment that loads `@deepseek-ai/dsh-tool-tasks`; loading the Task service alone is insufficient. SDK-generated spawn and fork compositions mount `@deepseek-ai/dsh-tasks-local` and `@deepseek-ai/dsh-tool-tasks` with the subagent tools. This dependency is the cost of using the same Task result, cancellation, and notification path for parent- and human-started activations.
|
||||
`TaskService.start()` rejects producers when no task controller is attached. A human-facing adapter that accepts child input must therefore attach a task controller, or run in a deployment that loads `@deepseek-ai/dsh-tool-tasks`; loading the Task service alone is insufficient. SDK-generated spawn and fork compositions mount `@deepseek-ai/dsh-tasks-local` and `@deepseek-ai/dsh-tool-tasks` with the subagent tools. This dependency is the cost of using the same Task result, cancellation, and notification path for parent- and human-started activations.
|
||||
|
||||
Cancellation always targets the whole current activation. If human and parent messages have joined one turn, either caller's cancellation aborts that turn, disposes its run, and settles its Task as `killed`; the messages do not have independent results or cancellation rights. `followup()` requires the caller's signal, and abort while live steering awaits request admission aborts the activation-owned controller so the provider discards the pending message and the call settles only after the child reaches quiescence. Independent cancellation requires a later message to start a separate turn instead of steering the current one.
|
||||
|
||||
@@ -121,7 +121,7 @@ Task records and active-run associations are process-local. Persistence makes th
|
||||
- Two callers may still race a stopped child through paths outside the continuation manager. The Agent registry prevents duplicate same-session publication; a losing Task fails and its message is not delivered. A message may also race cancellation, terminal status publication, or run disposal. Admission is not claimed to be atomic or exactly-once; the synchronous process-local association install closes duplicate cold resume through `followup` without a public lifecycle state machine.
|
||||
- Driving a continuable child through the ordinary Agent API bypasses its Task association. `ctx.subagents` rejects that live child as an ownership conflict; adapters must display persisted transcripts without loading an Agent and submit human input through `SubagentService.followup()`.
|
||||
- The active-run association coordinates only one runtime. Concurrent resume from multiple processes is not serialized; that deployment requires a persistence-level lease or compare-and-set operation.
|
||||
- Human interaction requires the exact parent Agent instance to remain live because owner disposal cancels and removes its Tasks. It also requires an attached Task control surface. Standalone child interaction requires a future separation between Task access ownership and durable notification targeting.
|
||||
- Human interaction requires the exact parent Agent instance to remain live because owner disposal cancels and removes its Tasks. It also requires an attached task controller. Standalone child interaction requires a future separation between Task access ownership and durable notification targeting.
|
||||
- The background tool returns child and Task ids before child publication and descriptor durability. Startup failure, a failed final durability confirmation, or process exit before the first child flush leaves the Task failed and may leave an unmaterialized or stale child id; by-id control reports missing state as unavailable rather than retroactively changing the tool acknowledgement.
|
||||
- Persisting explicit composition fields in the child log makes their lossless-JSON and compatibility contract part of resume. Later support for another composition input requires a deliberate descriptor-version change rather than implicitly persisting merge-extensible `AgentOptions` fields.
|
||||
- Task records and active-run associations are process-local even though child sessions are durable. Restart recovers the session, not in-flight work or its Task notification.
|
||||
|
||||
@@ -37,7 +37,7 @@ durable child Session
|
||||
|
||||
用户界面适配器打开 child 会话时，只读取持久化 transcript，不会仅为展示而恢复 agent。用户输入通过继续执行管理器，启动或加入与 parent 输入相同的 Task 激活。由用户启动的 Task 会保留当前加载的精确 parent Agent 作为通知目标，`task_output` 仍是唯一结果路径。只要 Task 尚未标记为已报告，现有完成监听器最多注入一条主动通知；`kill`、终态读取或终态等待都可能将其标记为已报告，并抑制这条通知。因此，仅允许在该 parent 实例保持存活时进行用户交互。可以比 parent 存活更久、并将结论显式合并回去的用户自有会话属于[交互式 side session](../../proposed/feature/2026-07-08-interactive-side-sessions.md)，不属于这一由 Task 持有的生命周期。
|
||||
|
||||
如果没有附加 Task 控制面，`TaskService.start()` 会拒绝 producer。因此，接受 child 输入的用户界面适配器必须附加 Task 控制面，或运行于加载了 `@deepseek-ai/dsh-tool-tasks` 的部署中；仅加载 Task 服务并不足够。SDK 生成的 spawn 与 fork 组合在挂载 subagent 工具时，也会挂载 `@deepseek-ai/dsh-tasks-local` 与 `@deepseek-ai/dsh-tool-tasks`。这项依赖是 parent 和用户启动的激活共用 Task 结果、取消和通知路径所付出的代价。
|
||||
如果没有附加任务控制器，`TaskService.start()` 会拒绝 producer。因此，接受 child 输入的用户界面适配器必须附加任务控制器，或运行于加载了 `@deepseek-ai/dsh-tool-tasks` 的部署中；仅加载 Task 服务并不足够。SDK 生成的 spawn 与 fork 组合在挂载 subagent 工具时，也会挂载 `@deepseek-ai/dsh-tasks-local` 与 `@deepseek-ai/dsh-tool-tasks`。这项依赖是 parent 和用户启动的激活共用 Task 结果、取消和通知路径所付出的代价。
|
||||
|
||||
取消始终作用于当前完整激活。如果用户消息和 parent 消息已经加入同一个轮次，任一调用方发起取消都会中止该轮次、dispose 其 run，并将对应 Task 结算为 `killed`；这些消息没有独立的结果或取消权。`followup()` 要求调用方提供信号；若在线 steering 正在等待请求准入时该信号被中止，激活自有的 controller 会被中止，以便提供方丢弃待处理消息，并且该调用仅在子 agent 完全停稳后结算。若需要独立取消，后续消息必须另起轮次，而不能加入当前轮次。
|
||||
|
||||
@@ -121,7 +121,7 @@ Task 记录和活跃 run 关联都位于进程内。持久化使 child 会话可
|
||||
- 两个调用方仍可能通过继续执行管理器外部的路径争抢已停止的 child。Agent 注册表会阻止相同会话的重复发布；失败的 Task 会失败，且其消息不会送达。消息也可能与取消、终态状态发布或 run dispose 发生竞态。准入不承诺原子或恰好执行一次；在进程内同步安装的关联无需公开生命周期状态机，即可通过 `followup` 消除重复的 cold resume。
|
||||
- 通过普通 Agent API 驱动可继续 child 会绕过其 Task 关联。`ctx.subagents` 会将该存活 child 视为所有权冲突并拒绝；适配器必须在不加载 Agent 的情况下展示持久化 transcript，并通过 `SubagentService.followup()` 提交用户输入。
|
||||
- 活跃 run 关联只能协调一个运行时。多个进程同时恢复时不会串行化；此类部署需要持久化层的租约或 compare-and-set 操作。
|
||||
- 用户交互要求作为 owner 的那个精确 parent Agent 实例保持存活，因为 dispose owner 会取消并移除其 Task。用户交互还要求附加 Task 控制面。若要单独与 child 交互，后续必须将 Task 访问所有权与持久化通知目标分离。
|
||||
- 用户交互要求作为 owner 的那个精确 parent Agent 实例保持存活，因为 dispose owner 会取消并移除其 Task。用户交互还要求附加任务控制器。若要单独与 child 交互，后续必须将 Task 访问所有权与持久化通知目标分离。
|
||||
- 后台工具会在 child 发布和描述符持久化之前返回 child id 和 Task id。启动失败、最终持久性确认失败，或进程在 child 首次 flush 之前退出，都会使 Task 失败，并可能留下 unmaterialized 或陈旧的 child id；按 id 的控制操作会将缺失状态报告为不可用，而不会追溯修改工具确认消息。
|
||||
- 将显式组合字段持久化到 child 日志后，其无损 JSON 与兼容性约定便成为恢复约定的一部分。后续如需支持其他组合配置输入，必须明确更改描述符版本，不能隐式持久化可通过声明合并扩展的 `AgentOptions` 字段。
|
||||
- Task 记录和活跃 run 关联位于进程内，而 child 会话具有持久性。重启会恢复会话，但不会恢复进行中的工作或其 Task 通知。
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-21-log-backed-session-titles.md
|
||||
2026-07-21-log-backed-session-titles.md: 47025ba7e72ea6746d6f4428a81d6d754a8dcd03
|
||||
2026-07-21-log-backed-session-titles.zh.md: d7e12875354ca1fc66f0035020c7ffe2ddaa9354
|
||||
2026-07-21-log-backed-session-titles.md: 46128a3e53d6617c9fa7a0c27d1ac01acfd1d9c1
|
||||
2026-07-21-log-backed-session-titles.zh.md: 2cdf5c22d6b2a28b27a68472619ced4b7dedd106
|
||||
|
||||
@@ -44,7 +44,7 @@ Automatic provider failures are nonfatal warnings and retain the latest title. E
|
||||
|
||||
A fork inherits seed title events unchanged, like the rest of its source log — a pinned (user-sourced) title stays pinned in the child until an explicit refresh. The first-message provider does not automatically retitle a fork. The all-messages provider may append a child-owned revision after a later child prompt, using inherited and new eligible messages.
|
||||
|
||||
`ctx.sessionQuery.readTitle()` folds one live-preferred or persisted log without loading titles during `listSessions()`. The TUI uses the latest title as its header subtitle and sets the terminal window title to `<session title> — <configured product title>` after terminal-safe rendering. The Web host folds the same log state into a validated mux control frame after each attached-session subscription baseline and immediately after forwarding a live raw title event. The browser retains only newer title event seqs even when the control frame precedes list or session-instance creation; sidebar labels, search, breadcrumbs, and the browser title then react to the projected revision. `session.list` remains metadata-only, so a cold persisted session uses the cwd basename or id until opening or resuming it attaches the log. The browser title uses `<session title> — <existing HTML title>` only for a selected titled session and otherwise preserves the product title. Consumers reporting agent completion use the core `findLastMessageTurnEnd()` fold, so a later between-turn title record cannot replace the preceding message-triggered outcome.
|
||||
`ctx.sessionQuery.readTitle()` folds one live-preferred or persisted log without loading titles during `listSessions()`. The TUI uses the latest title as its header subtitle and sets the terminal window title to `<session title> — <configured product title>` after terminal-safe rendering. The Web host folds the same log state into a validated mux control frame after each attached-session subscription baseline and immediately after forwarding a live raw title event. The browser retains only newer title event seqs even when the control frame precedes list or session-instance creation; sidebar labels, search, breadcrumbs, and the browser title then react to the projected revision. `session.list` remains metadata-only, so a cold persisted session uses the cwd basename or id until opening or resuming it attaches the log. The browser title uses `<session title> — <existing HTML title>` only for a selected titled session and otherwise preserves the product title. Consumers reporting agent completion use the core `foldConsumedWork()` fold, so a later between-turn title record cannot replace the preceding message-triggered outcome.
|
||||
|
||||
## Alternatives considered
|
||||
|
||||
|
||||
@@ -44,7 +44,7 @@ Status: implemented
|
||||
|
||||
与源日志的其他部分相同，fork 会原样继承作为种子的标题事件——被钉住（user 来源）的标题在子会话中保持钉住，直到显式 refresh。首消息提供方不会自动为 fork 重新生成标题。全部消息提供方可以在子会话出现后续提示词后追加一项归子会话所有的修订，并使用继承的合格消息和新增的合格消息。
|
||||
|
||||
`ctx.sessionQuery.readTitle()` 会折叠一份实时优先或已持久化的日志，而不会在 `listSessions()` 期间加载标题。TUI 使用最新标题作为其标题栏副标题，并在完成终端安全渲染后，将终端窗口标题设置为 `<session title> — <configured product title>`。Web host 会在每个已附加会话的订阅基线之后，以及转发实时原始标题事件后立即，将同一份日志状态折叠为经过校验的 mux 控制帧。即使控制帧先于列表或会话实例创建抵达，浏览器也只保留标题事件 seq 较新的版本；侧边栏标签、搜索、面包屑和浏览器标题会随投影后的修订更新。`session.list` 仍只包含元数据，因此尚未打开的持久化会话会继续以 cwd 基名或 id 作为回退，直至打开或恢复会话时附加其日志。浏览器仅在选中已有标题的会话时将标题设置为 `<session title> — <existing HTML title>`，否则保留产品标题。报告 agent 完成情况的消费方使用核心的 `findLastMessageTurnEnd()` 折叠逻辑，因此后续的轮次间标题记录无法取代此前由消息触发的结果。
|
||||
`ctx.sessionQuery.readTitle()` 会折叠一份实时优先或已持久化的日志，而不会在 `listSessions()` 期间加载标题。TUI 使用最新标题作为其标题栏副标题，并在完成终端安全渲染后，将终端窗口标题设置为 `<session title> — <configured product title>`。Web host 会在每个已附加会话的订阅基线之后，以及转发实时原始标题事件后立即，将同一份日志状态折叠为经过校验的 mux 控制帧。即使控制帧先于列表或会话实例创建抵达，浏览器也只保留标题事件 seq 较新的版本；侧边栏标签、搜索、面包屑和浏览器标题会随投影后的修订更新。`session.list` 仍只包含元数据，因此尚未打开的持久化会话会继续以 cwd 基名或 id 作为回退，直至打开或恢复会话时附加其日志。浏览器仅在选中已有标题的会话时将标题设置为 `<session title> — <existing HTML title>`，否则保留产品标题。报告 agent 完成情况的消费方使用核心的 `foldConsumedWork()` 折叠逻辑，因此后续的轮次间标题记录无法取代此前由消息触发的结果。
|
||||
|
||||
## 考虑过的替代方案
|
||||
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-22-durable-subagent-catalog-and-list-agents.md
|
||||
2026-07-22-durable-subagent-catalog-and-list-agents.md: 61bda2a52f2ea5db0a582fe668643a13b967f091
|
||||
2026-07-22-durable-subagent-catalog-and-list-agents.zh.md: 6b5bff468349df4226aaf76dd7ed7d9ce5c3f33c
|
||||
2026-07-22-durable-subagent-catalog-and-list-agents.md: 51664bc1e211802ef61f47508b9e57fdd099b9e7
|
||||
2026-07-22-durable-subagent-catalog-and-list-agents.zh.md: 79503a30e081abffd45637439ac539d95ae794f0
|
||||
|
||||
@@ -23,7 +23,7 @@ Parent-to-child enumeration is a service capability with consumer-specific proje
|
||||
- report corpus activity separately as `running` or `inactive`, without implying completion or resumability;
|
||||
- return every resulting child in stable `createdAt` ascending, child-id ascending order.
|
||||
|
||||
Every ordinary local start receives a `one-shot` descriptor with an optional caller-owned display label, while the continuation manager persists a labeled `continuable` descriptor containing its additional reconstruction fields. The model-facing delegation tool already owns a short `description` and supplies it for one-shot display; lower-level callers such as workflows need not invent presentation metadata. The model-facing `list_agents` adapter filters the service result to continuable children and refines status through the live Agent registry (`running`/`idle`/`complete`); a UI can consume both modes and choose an id-based fallback for unlabeled one-shot history. Descriptor persistence, by-id lookup, direct-parent authorization, and provider-independent cold resume remain owned by the implemented Activation contract. Listing consumes those facts but cannot weaken them or invent a second descriptor representation.
|
||||
Every ordinary local start receives a `one-shot` descriptor with an optional caller-owned display label, while the continuation manager persists a labeled `continuable` descriptor containing its additional reconstruction fields. The model-facing delegation tool already owns a short `description` and supplies it for one-shot display; lower-level callers such as workflows need not invent presentation metadata. The model-facing `list_agents` adapter filters the service result to continuable children and refines status through the live Agent registry (`running`/`idle`, and [`ready`](../bug-fix/2026-08-06-list-agents-residency-vocabulary.md) for storage-only); a UI can consume both modes and choose an id-based fallback for unlabeled one-shot history. Descriptor persistence, by-id lookup, direct-parent authorization, and provider-independent cold resume remain owned by the implemented Activation contract. Listing consumes those facts but cannot weaken them or invent a second descriptor representation.
|
||||
|
||||
### Enumeration decision
|
||||
|
||||
@@ -52,7 +52,7 @@ If measured scale later requires an index, that index is derived state: session
|
||||
|
||||
A valid descriptor produces one child entry, a per-child inspection failure produces one diagnostic entry, and a candidate without a descriptor produces no entry. `mode` is durable creation policy; `activity` is a process-local corpus snapshot. Activity is neither `AgentStatus`, the manager's internal Activation state, nor a durable outcome, and the result does not expose the internal `createdAt` sorting key. Exact Activation states and durable outcomes such as successful completion, failure, cancellation, and stop reason require a separate durable activation record and are outside this feature.
|
||||
|
||||
The model-facing `list_agents` tool takes one optional `scope: 'children' | 'descendants'` argument, derives the root id from the current execution Agent, and resolves the request through an explicit request-to-spec step (`undefined` → `children`) before either execution or rendering. The resolved `children` scope calls `SubagentService.listChildren(rootSessionId)`, while `descendants` calls `SubagentService.listDescendants(rootSessionId)`. Its internal output projection keeps `id` and `parent` as branded `SessionId` values until the tool JSON boundary. It keeps diagnostics, drops `one-shot` child entries, derives status from the live Agent registry — `running` for an active driver, `idle` for a resident Agent between turns, and `complete` when no live Agent remains — then renders `<id> [<status>] — <label>` or `<id> [diagnostic: <reason>]` in stable catalog order. The `descendants` scope flattens the complete tree from one live-preferred corpus in stable pre-order, traverses ordinary and one-shot intermediates so deeper continuable agents are discovered, revalidates each cold candidate against its enumerated lifecycle, and adds `parentId`/`depth` to every entry. The tool inserts ` parent=<id> depth=<n>` before the label; `parent` is the durable direct-parent session id and may name an omitted ordinary session. For the current caller, only depth-1 child rows are `send_message` candidates, while deeper child rows may be selected for `interrupt_agent` ([interrupt contract](2026-08-06-continuable-subagent-interrupt.md)). Discovery is a hint only — follow-up authority stays exact-direct-parent, and interrupt authority stays with the service's live-lineage check. An empty projection renders `(no subagents)`.
|
||||
The model-facing `list_agents` tool takes one optional `scope: 'children' | 'descendants'` argument, derives the root id from the current execution Agent, and resolves the request through an explicit request-to-spec step (`undefined` → `children`) before either execution or rendering. The resolved `children` scope calls `SubagentService.listChildren(rootSessionId)`, while `descendants` calls `SubagentService.listDescendants(rootSessionId)`. Its internal output projection keeps `id` and `parent` as branded `SessionId` values until the tool JSON boundary. It keeps diagnostics, drops `one-shot` child entries, derives status from the live Agent registry — `running` for an active driver, `idle` for a resident Agent between turns, and `ready` when no live Agent remains — resumable rather than terminal — then renders `<id> [<status>] — <label>` or `<id> [diagnostic: <reason>]` in stable catalog order. The `descendants` scope flattens the complete tree from one live-preferred corpus in stable pre-order, traverses ordinary and one-shot intermediates so deeper continuable agents are discovered, revalidates each cold candidate against its enumerated lifecycle, and adds `parentId`/`depth` to every entry. The tool inserts ` parent=<id> depth=<n>` before the label; `parent` is the durable direct-parent session id and may name an omitted ordinary session. For the current caller, only depth-1 child rows are `send_message` candidates, while deeper child rows may be selected for `interrupt_agent` ([interrupt contract](2026-08-06-continuable-subagent-interrupt.md)). Discovery is a hint only — follow-up authority stays exact-direct-parent, and interrupt authority stays with the service's live-lineage check. An empty projection renders `(no subagents)`.
|
||||
|
||||
In the superseded trace-based path, diagnostics used three fixed reasons. Malformed event surfaces, conflicting headers discovered during an exact child load, a read result whose immutable header differs from the traced candidate or no longer names the requested direct parent, a target that is no longer the located descriptor event, malformed descriptor content, and multiple descriptor events mapped to `corrupt`. An unknown descriptor version mapped to `unsupported`. `SESSION_QUERY_SESSION_NOT_FOUND`, `SESSION_QUERY_EVENT_NOT_FOUND`, and `SESSION_QUERY_PERSISTENCE_FAILED` from a per-child read mapped to `unavailable`. This phase boundary was intentional: a persistence outage during the initial trace failed the operation, while the same outage beginning during candidate reads could produce one identical `unavailable` diagnostic per affected child; the first version neither coalesced those diagnostics nor promoted them to a global failure. A missing descriptor was instead a non-subagent exclusion without a diagnostic. Configuration/window errors and unrecognized failures were not child diagnostics and propagated as operation failures. Each diagnostic identified the child id and reason without exposing model-hidden descriptor content; the candidate was omitted while healthy siblings remained visible. Sessions outside the trace's direct descendants were never read and produced no diagnostic.
|
||||
|
||||
@@ -95,7 +95,7 @@ The first version has no child deletion operation. If later product behavior del
|
||||
- `packages/subagent/subagent/tests/service.spec.ts` pins descriptor v2 parsing for both modes and proves an unlabeled raw start resolves a one-shot descriptor before provider dispatch. `packages/subagent/subagent-inprocess/tests/subagent-inprocess.spec.ts` proves the local driver appends that descriptor inside the initial turn, returns the published id when cancellation lands in the factory-to-run handoff, and keeps result and handle-disposal failures on separate channels. Delegation-tool tests pin propagation of their existing display description and preserve independent result and disposal diagnostics.
|
||||
- `packages/subagent/subagent/tests/list-children.spec.ts` pins the current read path against a real composition of the session store, JSONL persistence, spawn/fork providers, the subagent service, and the projection registry — no query service — keylessly: live-only listing without persistence; loud `SUBAGENT_CONTROL_PROJECTIONS_UNAVAILABLE` and `SUBAGENT_CONTROL_SESSION_STORE_UNAVAILABLE` even with zero children; the three-rung ladder (a live child never inspected, a cold child inspected exactly once, and the cache-hit, absent-key, absent-service, and poisoned-row second-rung cases); last-wins over multiple descriptors; malformed payloads and unknown versions diagnosed as `corrupt`; a failed cold inspection as one `unavailable` diagnostic retried on the next listing; a fork seed's ancestor descriptor listed under that identity; foreign-unit fold failures contained per child as `corrupt` on both the live and cold paths; `createdAt`-then-id ordering without ordinary forks; provider absence without child omission; compacted/uncompacted twins listing identically; a persisted-listing failure failing the whole enumeration; cancellation normalized to stable `CANCELLED`; typed stable error codes; and descendant listing's iterative stable pre-order, traversal through ordinary and one-shot intermediates, positioned diagnostics, lifecycle revalidation, and cancellation. A companion spec (retired together with the query-backed read path) rejected eager evaluation of the optional session-query runtime while importing the ordinary subagent surface.
|
||||
- `packages/subagent/tool-subagent-control/tests/list-agents.spec.ts` pins the `list_agents` schema (one optional `scope` enum), the continuable-only projection that omits a healthy one-shot sibling while preserving diagnostics, registry-derived child/diagnostic/empty text forms, an end-to-end settled-child listing with its durable label, the descendants scope's pre-order parent/depth annotations across a live waiting branch, cancellation forwarding to both scopes, the no-agent rejection, the `agents` load requirement without `sessionQuery`, and HMR disposal.
|
||||
- The keyless ACP snapshot scenario `subagent-list-agents` (examples/acp-agent) fences its second parent turn on a snapshot-only `subagent/end` marker, then executes `list_agents` for real against the subagent service, the projection registry, and JSONL persistence, rendering `<id> [complete] — <label>`.
|
||||
- The keyless ACP snapshot scenario `subagent-list-agents` (examples/acp-agent) fences its second parent turn on a snapshot-only `subagent/end` marker, then executes `list_agents` for real against the subagent service, the projection registry, and JSONL persistence, rendering `<id> [ready] — <label>`.
|
||||
- The keyless snapshot scenario `subagent-diagnostic` (examples/headless-agent) pins the current listing's model-visible diagnostic classification, including a descriptor-less settled child surfacing as a `corrupt` diagnostic.
|
||||
- The keyless ACP snapshot scenario `subagent-published-run-failure` publishes a real one-shot child, injects independent run-result and handle-disposal failures, and preserves both diagnostics in the parent tool result.
|
||||
|
||||
|
||||
@@ -23,7 +23,7 @@ parent 到 child 的枚举是一项带消费方专用投影的服务功能。`Su
|
||||
- 将语料活动状态单独报告为 `running` 或 `inactive`，但不暗示已完成或可恢复；
|
||||
- 按 `createdAt` 升序、再按 child id 升序稳定返回所有结果 child。
|
||||
|
||||
每次普通的本地启动都会收到带可选、由调用方拥有之显示标签的 `one-shot` 描述符，而继续执行管理器会持久化带标签、包含附加重建字段的 `continuable` 描述符。面向模型的委派工具已经拥有简短 `description`，会将其用于一次性显示；workflow 等底层调用方无需凭空构造展示元数据。面向模型的 `list_agents` 适配器会将服务结果过滤为可继续 child，并通过在线 Agent 注册表细化状态（`running`／`idle`／`complete`）；UI 可以消费两种模式，并为无标签的一次性历史选择基于 id 的回退展示。描述符持久化、按 id 查找、直接 parent 鉴权和不依赖提供方的冷恢复仍归已实现的 Activation 约定负责。列表查询消费这些事实，但不能削弱它们，也不能另行发明第二种描述符表示。
|
||||
每次普通的本地启动都会收到带可选、由调用方拥有之显示标签的 `one-shot` 描述符，而继续执行管理器会持久化带标签、包含附加重建字段的 `continuable` 描述符。面向模型的委派工具已经拥有简短 `description`，会将其用于一次性显示；workflow 等底层调用方无需凭空构造展示元数据。面向模型的 `list_agents` 适配器会将服务结果过滤为可继续 child，并通过在线 Agent 注册表细化状态（`running`／`idle`，以及对应仅存于存储的 [`ready`](../bug-fix/2026-08-06-list-agents-residency-vocabulary.md)）；UI 可以消费两种模式，并为无标签的一次性历史选择基于 id 的回退展示。描述符持久化、按 id 查找、直接 parent 鉴权和不依赖提供方的冷恢复仍归已实现的 Activation 约定负责。列表查询消费这些事实，但不能削弱它们，也不能另行发明第二种描述符表示。
|
||||
|
||||
### 枚举决策
|
||||
|
||||
@@ -52,7 +52,7 @@ subagent 服务将 `sessionQuery` 保持为可选依赖，因此没有该服务
|
||||
|
||||
有效描述符产生一个 child 条目，逐 child 检查失败产生一个 diagnostic 条目，缺少描述符的候选不产生条目。`mode` 是持久化创建策略；`activity` 是进程本地语料快照。活动状态既不是 `AgentStatus`、管理器内部的 Activation 状态，也不是持久化结果，结果不公开内部 `createdAt` 排序键。成功完成、失败、取消和停止原因等精确 Activation 状态与持久化结果需要单独的持久化激活记录，不在本功能范围内。
|
||||
|
||||
面向模型的 `list_agents` 工具接受一个可选的 `scope: 'children' | 'descendants'` 参数，从当前执行 Agent 推导根 id，并在执行或渲染前通过显式的 request-to-spec 步骤解析请求（`undefined` → `children`）。解析后的 `children` scope 调用 `SubagentService.listChildren(rootSessionId)`，`descendants` scope 则调用 `SubagentService.listDescendants(rootSessionId)`。其内部输出投影中的 `id` 与 `parent` 会一直保持为品牌化的 `SessionId` 值，直到工具 JSON 边界。它保留 diagnostic，丢弃 `one-shot` child 条目，状态取自在线 Agent 注册表——driver 活跃为 `running`，驻留但处于轮次之间为 `idle`，没有在线 Agent 时为 `complete`——然后按稳定目录顺序渲染 `<id> [<status>] — <label>` 或 `<id> [diagnostic: <reason>]`。`descendants` scope 从一份实时优先语料按稳定 pre-order 展平完整树，遍历普通与一次性中间节点以发现更深的可继续 agent，依据枚举生命周期重新校验每个冷候选，并为每个条目附加 `parentId`／`depth`。工具会在 label 之前插入 ` parent=<id> depth=<n>`；`parent` 是持久化直接 parent 会话 id，可能指向被省略的普通会话。对于当前调用方，只有 depth-1 child 条目可作为 `send_message` 候选，更深的 child 条目则可供 `interrupt_agent` 选择（[中断约定](2026-08-06-continuable-subagent-interrupt.md)）。发现结果只是提示——follow-up 权限仍仅属于确切直接 parent，中断权限仍由服务的在线 lineage 检查决定。空投影渲染为 `(no subagents)`。
|
||||
面向模型的 `list_agents` 工具接受一个可选的 `scope: 'children' | 'descendants'` 参数，从当前执行 Agent 推导根 id，并在执行或渲染前通过显式的 request-to-spec 步骤解析请求（`undefined` → `children`）。解析后的 `children` scope 调用 `SubagentService.listChildren(rootSessionId)`，`descendants` scope 则调用 `SubagentService.listDescendants(rootSessionId)`。其内部输出投影中的 `id` 与 `parent` 会一直保持为品牌化的 `SessionId` 值，直到工具 JSON 边界。它保留 diagnostic，丢弃 `one-shot` child 条目，状态取自在线 Agent 注册表——driver 活跃为 `running`，驻留但处于轮次之间为 `idle`，没有在线 Agent 时为 `ready`（可恢复而非终态）——然后按稳定目录顺序渲染 `<id> [<status>] — <label>` 或 `<id> [diagnostic: <reason>]`。`descendants` scope 从一份实时优先语料按稳定 pre-order 展平完整树，遍历普通与一次性中间节点以发现更深的可继续 agent，依据枚举生命周期重新校验每个冷候选，并为每个条目附加 `parentId`／`depth`。工具会在 label 之前插入 ` parent=<id> depth=<n>`；`parent` 是持久化直接 parent 会话 id，可能指向被省略的普通会话。对于当前调用方，只有 depth-1 child 条目可作为 `send_message` 候选，更深的 child 条目则可供 `interrupt_agent` 选择（[中断约定](2026-08-06-continuable-subagent-interrupt.md)）。发现结果只是提示——follow-up 权限仍仅属于确切直接 parent，中断权限仍由服务的在线 lineage 检查决定。空投影渲染为 `(no subagents)`。
|
||||
|
||||
在已被取代的追踪读路径中，diagnostic 使用三种固定原因。格式错误的事件 surface、精确加载 child 时发现的 header 冲突、读取结果中的不可变 header 与追踪到的候选不一致或不再指向请求的直接 parent、读取目标不再是先前定位的描述符事件、格式错误的描述符内容和多个描述符事件映射为 `corrupt`。未知描述符版本映射为 `unsupported`。逐 child 读取产生的 `SESSION_QUERY_SESSION_NOT_FOUND`、`SESSION_QUERY_EVENT_NOT_FOUND` 和 `SESSION_QUERY_PERSISTENCE_FAILED` 映射为 `unavailable`。这项阶段边界是有意为之：初始追踪期间发生持久化故障会让操作失败，而同一故障如果始于候选读取期间，可能会让每个受影响的 child 分别产生一条相同的 `unavailable` diagnostic；第一版既不合并这些 diagnostic，也不会把它们提升为全局失败。缺少描述符则作为非 subagent 排除，且不产生 diagnostic。配置错误、窗口错误和未识别的失败不属于 child diagnostic，会作为操作失败继续向上传播。每条 diagnostic 都标识 child id 及原因，不暴露对模型隐藏的描述符内容；系统会排除该候选，而其他健康的 sibling 仍然可见。系统绝不会读取不属于追踪结果直接后代的会话，也不会为它们产生 diagnostic。
|
||||
|
||||
@@ -95,7 +95,7 @@ diagnostic 是瞬时查询结果，不属于会话事件或目录状态。推导
|
||||
- `packages/subagent/subagent/tests/service.spec.ts` 固定两种模式下的描述符 v2 解析，并证明无标签的底层启动会在分发给提供方之前解析出一次性描述符。`packages/subagent/subagent-inprocess/tests/subagent-inprocess.spec.ts` 证明本地驱动会在初始轮次内追加该描述符，在取消落入工厂到 run 的交接窗口时返回已发布 id，并让结果与句柄释放失败保留在独立通道中。委派工具测试固定其现有显示说明的传递，并保留相互独立的结果与 dispose diagnostic。
|
||||
- `packages/subagent/subagent/tests/list-children.spec.ts` 针对由会话存储、JSONL 持久化、spawn/fork 提供方、subagent 服务与投影注册表构成的真实组合——不含查询服务——以无密钥方式钉住现行读取路径：无持久化时的仅存活列表；零 children 也响亮报 `SUBAGENT_CONTROL_PROJECTIONS_UNAVAILABLE` 与 `SUBAGENT_CONTROL_SESSION_STORE_UNAVAILABLE`；三级阶梯（存活 child 从不检查、冷 child 恰好检查一次，以及缓存命中、key 缺席、服务缺席、行中毒四个第二级用例）；多描述符 last-wins 取末者；载荷格式错误与未知版本诊断为 `corrupt`；冷检查失败成一条 `unavailable` diagnostic 并在下次列表重试；fork seed 中的祖先描述符按该身份列出；外部 unit 折叠失败在存活与冷两条路径上按 child 收纳为 `corrupt`；按 `createdAt` 再按 id 排序且不列普通 fork；提供方缺失时不排除 child；压缩与未压缩的孪生 child 列表结果一致；持久化列表失败使整次枚举失败；取消稳定归一化为 `CANCELLED`；带类型的稳定错误码；以及后代列表的迭代式稳定 pre-order、穿过普通与一次性中间节点、带位置 diagnostic、生命周期复验与取消。一个伴随规格(已随查询式读取路径一起退役)曾在导入普通 subagent surface 时拒绝对可选 session-query 运行时的 eager 求值。
|
||||
- `packages/subagent/tool-subagent-control/tests/list-agents.spec.ts` 固定 `list_agents` 的 schema（一个可选 `scope` 枚举）、只保留可继续 child 且排除健康的一次性 sibling、同时保留 diagnostic 的投影、由注册表推导的 child／diagnostic／空结果文本形式、带持久化 label 的已结束 child 端到端列表、descendants scope 在在线 waiting 分支上的 pre-order parent/depth 注释、两个 scope 的取消信号转发、无调用 agent 时的拒绝、要求 `agents` 但不再注入 `sessionQuery` 的加载约定，以及 HMR dispose。
|
||||
- 无密钥 ACP 快照场景 `subagent-list-agents`（examples/acp-agent）使用仅限快照的 `subagent/end` 标记为第二个 parent 轮次设置边界，随后针对 subagent 服务、投影注册表和 JSONL 持久化真实执行 `list_agents`，渲染 `<id> [complete] — <label>`。
|
||||
- 无密钥 ACP 快照场景 `subagent-list-agents`（examples/acp-agent）使用仅限快照的 `subagent/end` 标记为第二个 parent 轮次设置边界，随后针对 subagent 服务、投影注册表和 JSONL 持久化真实执行 `list_agents`，渲染 `<id> [ready] — <label>`。
|
||||
- 无密钥快照场景 `subagent-diagnostic`（examples/headless-agent）钉住现行列表的模型可见诊断分类，包括无描述符的定局 child 以 `corrupt` diagnostic 出现。
|
||||
- 无密钥 ACP 快照场景 `subagent-published-run-failure` 会发布一个真实的一次性 child，注入相互独立的 run result 与 handle dispose 失败，并在 parent 工具结果中保留两项 diagnostic。
|
||||
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-22-web-multimodal-image-input-and-durable-attachments.md
|
||||
2026-07-22-web-multimodal-image-input-and-durable-attachments.md: 2a3925d8b779fed7f8b3330d079fdb3059685fd5
|
||||
2026-07-22-web-multimodal-image-input-and-durable-attachments.md: ea234e557553dda03ef2b707b60d47d3f43eb8f8
|
||||
2026-07-22-web-multimodal-image-input-and-durable-attachments.zh.md: be478ece73039201a820d804435610faea4cb3fe
|
||||
|
||||
@@ -128,7 +128,7 @@ The Pi-AI adapter is the first visual-input route: it resolves `ctx.attachments`
|
||||
|
||||
Core supports structured assistant image blocks, but no current production provider route is certified for image output. Any future output-capable adapter must retrieve provider bytes under bounded size and time policy, validate them through the same attachment service, persist them, and only then publish the atomic `ImageBlock`. A URL in assistant Markdown remains text and is never downloaded automatically.
|
||||
|
||||
Provider-neutral token estimation does not guess visual pricing from image dimensions; provider-reported usage remains authoritative. ACP renders an explicit image marker until that protocol surface gains native image support rather than silently omitting the block.
|
||||
Provider-neutral token estimation does not guess visual pricing from image dimensions; provider-reported usage remains authoritative. ACP renders an explicit image marker until that protocol API gains native image support rather than silently omitting the block.
|
||||
|
||||
Compaction replays the selected conversation prefix, including image references, into the configured summarization route. A visual-capable route resolves those references through its adapter; a text-only route fails explicitly instead of silently dropping the visual context. The synthesized checkpoint remains text-only, and `compact-basic` rejects image summary output with `UNSUPPORTED_CONTENT`.
|
||||
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-23-web-assistant-markdown.md
|
||||
2026-07-23-web-assistant-markdown.md: ad86559e3b5294b6bd67d69ff5c6ce37a5172008
|
||||
2026-07-23-web-assistant-markdown.md: 0ad74546b9a54d5eadfc3e7991efa4e0d9b7bc77
|
||||
2026-07-23-web-assistant-markdown.zh.md: e7544b2c731581e29b026026a2bb627defb91ca7
|
||||
|
||||
@@ -28,7 +28,7 @@ Fenced code and GFM tables own horizontal overflow so long content cannot widen
|
||||
|
||||
**Promote the existing mdast and micromark development dependencies and maintain a custom React walker.** This avoids a new parser family but makes the product own every node mapping, GFM extension, and security-sensitive rendering branch. The dedicated React renderer keeps that traversal upstream while preserving an AST-to-React path. *Later reversed on new evidence — incremental streaming parsing needs AST-level input the string-only wrapper cannot provide; the [incremental AST renderer note](../architecture/2026-08-06-web-markdown-incremental-ast-renderer.md) owns that decision.*
|
||||
|
||||
**Replace `MessageText` with Markdown rendering.** This formats user prompts and steering as a side effect. Those authored surfaces remain literal until the product chooses that behavior explicitly.
|
||||
**Replace `MessageText` with Markdown rendering.** This formats user prompts and steering as a side effect. Those authored inputs remain literal until the product chooses that behavior explicitly.
|
||||
|
||||
**Parse Markdown into session snapshots.** This would make React nodes or presentation ASTs durable runtime state and reintroduce a final-versus-streaming mode boundary. Parsing stays at the presentation leaf instead.
|
||||
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-28-skill-invocation-policy.md
|
||||
2026-07-28-skill-invocation-policy.md: 67b50c7f92f7c93a13b495877bb8f87cfac44552
|
||||
2026-07-28-skill-invocation-policy.md: 7a4f83ecae3dea82ab6864e748c9d8b8599f6bfc
|
||||
2026-07-28-skill-invocation-policy.zh.md: 20d41ec1b2fdbfe5bbbe3c54a681dd4f807c703e
|
||||
|
||||
@@ -22,7 +22,7 @@ The model-facing `dsh-tool-skill` catalog and loader enforce `isModelInvocable`.
|
||||
|
||||
These rules permit all four combinations:
|
||||
|
||||
| Policy | Model surface | User surface |
|
||||
| Policy | Model invocation | User invocation |
|
||||
|---|---|---|
|
||||
| `{ modelInvocable: true, userInvocable: true }` | included | included |
|
||||
| `{ modelInvocable: true, userInvocable: false }` | included | excluded |
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-07-30-continuable-subagent-report-tool.md
|
||||
2026-07-30-continuable-subagent-report-tool.md: 6a607f1da48fdfde36b2d9351f131dd3ef67ba7f
|
||||
2026-07-30-continuable-subagent-report-tool.zh.md: b4d65fc71695e9fca46f4e2c69c9749745de260a
|
||||
2026-07-30-continuable-subagent-report-tool.md: d4681d0ea30335a5537525927ecad615e6edd8b8
|
||||
2026-07-30-continuable-subagent-report-tool.zh.md: 731225d2a5b788997704fafe9331901ca9b84393
|
||||
|
||||
@@ -12,7 +12,7 @@ Treating every final assistant message as an implicit result would conflate turn
|
||||
|
||||
## Decision
|
||||
|
||||
Add the independently installed `@deepseek-ai/dsh-tool-subagent-report` package. It contributes an ordinary model-facing `report` tool to each continuable in-process child Activation. A child may call it zero or multiple times in a turn. Success neither concludes the turn, settles the Activation, nor prevents later parent follow-ups, and finishing a turn never reports automatically.
|
||||
Add the independently installed `@deepseek-ai/dsh-tool-subagent-report` package. It contributes an ordinary model-facing `report` tool to each continuable in-process child Activation. The mechanism accepts zero or multiple calls in a turn; the child is separately instructed to call it once before finishing ([the report obligation](2026-08-06-continuable-child-report-obligation.md)). Success neither concludes the turn, settles the Activation, nor prevents later parent follow-ups, and finishing a turn never reports automatically.
|
||||
|
||||
The feature is a collaboration control, not a result-bearing execution wrapper. It adds no Task, `SubagentRun`, result promise, Activation state, delivery queue, or replay path.
|
||||
|
||||
@@ -22,7 +22,7 @@ The feature is a collaboration control, not a result-bearing execution wrapper.
|
||||
|
||||
`messageId` is the stable `MessageId` of the user-role message accepted by the parent. It is not an `InboxItemId`: quiet delivery creates no inbox occurrence, while waking delivery creates one occurrence for the same stable message. It is also not a read receipt, parent-log acknowledgement, turn-completion receipt, or persistence flush.
|
||||
|
||||
The description states that reporting is explicit, repeatable, direct-parent-only, and non-terminal. It warns that a failed tool result may still follow an accepted send because a later `tools/post-execute` failure can replace the result. Without an idempotency key, stronger wording would encourage duplicate retries after ambiguous failure.
|
||||
The description states that reporting is required before finishing, repeatable, direct-parent-only, and non-terminal. It warns that a failed tool result may still follow an accepted send because a later `tools/post-execute` failure can replace the result. Without an idempotency key, stronger wording would encourage duplicate retries after ambiguous failure.
|
||||
|
||||
The tool uses generic rendering with no locations. Its acknowledgement includes `messageId`. Scope-local registration keeps presentation and execution aligned: roots, one-shot children, remote providers, sibling scopes, and agentless execution neither see nor execute `report`. It installs after the child's global `toolFilter`, so a delegation allow-list cannot accidentally remove the structural return channel; deployments that require no return channel omit the package.
|
||||
|
||||
@@ -36,7 +36,7 @@ Nested reporting crosses exactly one edge. A grandchild reports to its direct ch
|
||||
|
||||
### Delivery policy
|
||||
|
||||
The package validates `reportDelivery: 'quiet' | 'wakeup'`; the default is `quiet`.
|
||||
The package validates `reportDelivery: 'quiet' | 'wakeup'`; the default is `wakeup` ([why the default reversed](2026-08-06-continuable-child-report-obligation.md)).
|
||||
|
||||
Quiet delivery calls `parent.inject()`. It adds model-visible context without starting a parent model request: an idle parent appends before the call returns, while an admitting or running parent stages the report for the next safe log position. It creates no inbox occurrence and therefore no synthetic continuation-manager acceptance record.
|
||||
|
||||
@@ -56,13 +56,13 @@ The subagent seam adds `registerContinuableSetup(contribution): () => void`, bac
|
||||
|
||||
The registry owns registration, per-child installation records, setup rollback, child-scope cleanup, and immediate revocation. Applying a batch returns the Agent setup commit that revalidates provisioning after every setup await and immediately before Agent publication. A throwing or concurrently revoked contribution therefore rejects before either Agent or Session publication and rolls back the batch. New registrations affect a resident child only on its next Activation; removing a registration first closes it to new setup and then revokes every provisioning or resident installation immediately. Registration disposal and child-context disposal are idempotent and attempt every release before aggregating failures.
|
||||
|
||||
This seam keeps the continuation manager unaware of tool names. The report package installs only `report`; `@deepseek-ai/dsh-tool-subagent-control` independently installs parent-side `send_message` and `list_agents`. A deployment can install either direction, both, or neither. Providers remain data-only, durable descriptors do not snapshot report availability or delivery mode, and cold resume uses the deployment's current contributions and policy.
|
||||
This seam keeps the continuation manager unaware of tool names. The report package installs only `report` and its child-scoped guidance section; `@deepseek-ai/dsh-tool-subagent-control` independently installs parent-side `send_message` and `list_agents`. A deployment can install either direction, both, or neither. Providers remain data-only, durable descriptors do not snapshot report availability or delivery mode, and cold resume uses the deployment's current contributions and policy.
|
||||
|
||||
### Snapshot coverage
|
||||
|
||||
The ACP snapshot harness adds `waitForSubagentTurnEnd`, selecting the Nth harvested child by the same order as `session.N.jsonl`. It waits for a closed child turn containing a request header so a continuable child's earlier descriptor-seed turn cannot satisfy the boundary. This lets the assembled quiet-mode scenario wait for the child-side report without inventing a parent-visible signal.
|
||||
The ACP snapshot harness adds `waitForSubagentTurnEnd`, selecting the Nth harvested child by the same order as `session.N.jsonl`. It waits for a closed child turn containing a request header so a continuable child's earlier descriptor-seed turn cannot satisfy the boundary. This lets the assembled scenario wait for the child-side report without inventing a parent-visible signal.
|
||||
|
||||
The authored snapshot starts a continuable child, executes the real scope-local `report` tool, confirms that the idle parent is not woken, and then submits a later parent prompt that consumes the framed report. It declares child schema pin `1`, so the otherwise non-global `report` schema is checked against `tool-schemas.1.expected.json` while the root keeps the default schema pin. The generated tool catalog separately mints a child scope to include the same scope-local schema.
|
||||
The authored snapshot starts a continuable child, executes the real scope-local `report` tool, observes the one ordinary parent turn the default waking delivery creates, and then submits a later parent prompt that consumes the framed report. It declares child pins `1`, so the otherwise non-global `report` schema and the child's own prompt are checked against `tool-schemas.1.expected.json` and `system-prompt.1.expected.md` while the root keeps the class pins. The generated tool catalog separately mints a child scope to include the same scope-local schema.
|
||||
|
||||
## Alternatives considered
|
||||
|
||||
@@ -72,7 +72,7 @@ Automatic delivery cannot represent zero reports, progress reports, or several s
|
||||
|
||||
### Always wake the parent
|
||||
|
||||
Waking on every report creates unsolicited turns and can cascade through nested subagents. Quiet delivery matches background coordination better as the default, while deployments that require immediate action can select wakeup.
|
||||
Waking on every report creates unsolicited turns and can cascade through nested subagents. Quiet delivery was chosen as the default on the assumption that the parent had another reason to read its context. [The report obligation](2026-08-06-continuable-child-report-obligation.md) supersedes that choice: a parked background coordinator has no such reason, so waking is the default and this paragraph now records why `quiet` still exists.
|
||||
|
||||
### Let the child choose the delivery mode
|
||||
|
||||
@@ -103,16 +103,16 @@ A post-creation revocation check can reject the Activation only after the Agent
|
||||
- A continuable in-process child exposes exactly one scope-local `report` schema only while the report package's contribution is installed; unrelated Agents never expose it.
|
||||
- The tool returns the parent message's stable `MessageId`. Quiet delivery has no `InboxItemId`; waking delivery has a separate inbox occurrence.
|
||||
- Only the exact resident child may report, and only to the exact live direct parent derived from durable lineage. The service has no recipient parameter or offline fallback.
|
||||
- Quiet delivery is the validated default and never starts a parent request. Wakeup creates exactly one later FIFO turn and never steers an open turn.
|
||||
- Waking delivery is the validated default: it creates exactly one later FIFO turn and never steers an open turn. Quiet delivery never starts a parent request.
|
||||
- Child cancellation or disposal after parent acceptance does not retract the report. Before acceptance, child disposal, drain, parent loss, or caller cancellation rejects the operation.
|
||||
- Fresh and resumed Activations compose current setup contributions before publication. Grants wait for the next Activation; revocation is immediate for resident children.
|
||||
- Unit coverage pins visibility, allow-list behavior, both delivery modes, stable message and sender identities, nested routing, invalid senders, absent parents, cancellation, drain, revocation races, and the absence of Tasks or implicit final reporting.
|
||||
- The keyless assembled snapshot proves the real child tool, quiet non-wakeup behavior, durable parent framing, and later parent consumption.
|
||||
- The keyless assembled snapshot proves the real child tool, the one waking parent turn, durable parent framing, and later parent consumption.
|
||||
|
||||
### Accepted risks
|
||||
|
||||
The acceptance boundary is weaker than durable end-to-end delivery. A crash can leave the result ambiguous, and retries may duplicate reports.
|
||||
|
||||
Wakeup mode can amplify model work when nested children report frequently. Deployment ownership and a quiet default limit but do not remove that risk.
|
||||
Waking delivery can amplify model work when nested children report frequently. Deployment ownership through `reportDelivery` bounds but does not remove that risk.
|
||||
|
||||
Registry presence is the parent liveness signal. A host-owned parent whose `AgentHandle.dispose()` has started but has not yet unwound its scope can still accept and append a report that it will not act on in this process. Closing that gap requires an Agent-level disposal-start signal rather than subagent-layer inference.
|
||||
|
||||
@@ -12,7 +12,7 @@ Status: implemented
|
||||
|
||||
## 决策
|
||||
|
||||
新增可独立安装的 `@deepseek-ai/dsh-tool-subagent-report` 包。它会向每个可继续进程内 child Activation 贡献一个普通的面向模型 `report` 工具。child 在一个轮次中可调用零次或多次。调用成功既不会结束该轮次或结算 Activation，也不会阻止 parent 之后继续 follow-up；完成轮次也绝不会自动报告。
|
||||
新增可独立安装的 `@deepseek-ai/dsh-tool-subagent-report` 包。它会向每个可继续进程内 child Activation 贡献一个普通的面向模型 `report` 工具。机制本身接受一个轮次中调用零次或多次；child 会另行被要求在结束前调用一次（见[报告义务](2026-08-06-continuable-child-report-obligation.md)）。调用成功既不会结束该轮次或结算 Activation，也不会阻止 parent 之后继续 follow-up；完成轮次也绝不会自动报告。
|
||||
|
||||
该功能是协作控制，不是承载结果的执行包装层。它不新增 Task、`SubagentRun`、结果 promise、Activation 状态、投递队列或回放路径。
|
||||
|
||||
@@ -22,7 +22,7 @@ Status: implemented
|
||||
|
||||
`messageId` 是 parent 接受的用户角色消息所对应的稳定 `MessageId`。它不是 `InboxItemId`：静默投递不创建 inbox 条目实例，唤醒投递则会为同一条稳定消息创建一个条目实例。它也不是已读回执、parent 日志确认、轮次完成回执或持久化 flush。
|
||||
|
||||
工具描述会明确报告操作是显式、可重复、仅限直接 parent 且不会结束轮次的。它还会警告：发送被接受后，后续 `tools/post-execute` 失败可能替换工具结果，因此工具结果失败时内容仍可能已经送达。没有幂等键时，更强的表述会诱导调用方在结果不明确的失败后重复重试。
|
||||
工具描述会明确报告操作在结束前必须执行、可重复、仅限直接 parent 且不会结束轮次。它还会警告：发送被接受后，后续 `tools/post-execute` 失败可能替换工具结果，因此工具结果失败时内容仍可能已经送达。没有幂等键时，更强的表述会诱导调用方在结果不明确的失败后重复重试。
|
||||
|
||||
该工具使用不带 location 的通用渲染，其确认中包含 `messageId`。作用域局部注册使呈现与执行保持一致：root、one-shot child、远程提供方、同级作用域和无 agent（智能体）执行既不能看到，也不能执行 `report`。它会在 child 的全局 `toolFilter` 之后安装，因此委派 allow-list 不会意外移除这条结构性返回通道；不需要返回通道的部署不安装该包。
|
||||
|
||||
@@ -36,7 +36,7 @@ root、one-shot child、伪造对象、陈旧 Agent 和同 id 替换对象都以
|
||||
|
||||
### 投递策略
|
||||
|
||||
该包会校验 `reportDelivery: 'quiet' | 'wakeup'`，默认值为 `quiet`。
|
||||
该包会校验 `reportDelivery: 'quiet' | 'wakeup'`，默认值为 `wakeup`（见[默认值反转的理由](2026-08-06-continuable-child-report-obligation.md)）。
|
||||
|
||||
静默投递调用 `parent.inject()`。它会添加模型可见上下文，但不启动 parent 模型请求：若 parent 空闲，则在调用返回前追加消息；若 parent 正在准入或运行，则暂存报告，留到下一个安全日志位置。该模式不创建 inbox 条目实例，因此也不会产生虚构的继续执行管理器接受记录。
|
||||
|
||||
@@ -56,13 +56,13 @@ subagent seam 新增 `registerContinuableSetup(contribution): () => void`，由
|
||||
|
||||
注册表负责注册、每个 child 的安装记录、设置回滚、child 作用域清理和立即撤销。应用一个批次会返回 Agent setup 提交对象，用于在所有 setup 的 await 均结算后、紧邻 Agent 发布前重新校验配置状态。因此，某项贡献抛出异常或被并发撤销时，会在 Agent 与会话发布前拒绝操作并回滚该批次。新注册项只会在驻留 child 的下一个 Activation 生效；移除注册项时，会先将它对新设置关闭，再立即撤销为正在配置或驻留的每个 child 安装的实例。注册 dispose（资源释放）与 child 上下文 dispose 都是幂等的，两者都会先尝试每项释放，再聚合失败。
|
||||
|
||||
该 seam 使继续执行管理器无需知道工具名。report 包只安装 `report`；`@deepseek-ai/dsh-tool-subagent-control` 则独立安装 parent 侧的 `send_message` 和 `list_agents`。部署时可安装任一方向、同时安装两者或两者均不安装。提供方仍只负责数据，持久化描述符不会对 report 可用性或投递模式建立快照，冷恢复则使用部署当前的贡献与策略。
|
||||
该 seam 使继续执行管理器无需知道工具名。report 包只安装 `report` 及其 child 作用域指引 section；`@deepseek-ai/dsh-tool-subagent-control` 则独立安装 parent 侧的 `send_message` 和 `list_agents`。部署时可安装任一方向、同时安装两者或两者均不安装。提供方仍只负责数据，持久化描述符不会对 report 可用性或投递模式建立快照，冷恢复则使用部署当前的贡献与策略。
|
||||
|
||||
### 快照覆盖
|
||||
|
||||
ACP（Agent Client Protocol）快照 harness 新增 `waitForSubagentTurnEnd`，按与 `session.N.jsonl` 相同的顺序选择第 N 个已收集 child。它会等待一个包含请求 header 的已闭合 child 轮次，以防可继续 child 早期播种描述符的轮次错误满足该边界。这样，整体组装的静默模式场景无需伪造 parent 可见信号，就能等待 child 侧报告。
|
||||
ACP（Agent Client Protocol）快照 harness 新增 `waitForSubagentTurnEnd`，按与 `session.N.jsonl` 相同的顺序选择第 N 个已收集 child。它会等待一个包含请求 header 的已闭合 child 轮次，以防可继续 child 早期播种描述符的轮次错误满足该边界。这样，整体组装的场景无需伪造 parent 可见信号，就能等待 child 侧报告。
|
||||
|
||||
手写快照会启动一个可继续 child，执行真实的作用域局部 `report` 工具，确认空闲 parent 未被唤醒，然后提交一条后续 parent 提示词，使其消费封装后的报告。它声明 child schema pin `1`，因此本不属于全局的 `report` schema 会与 `tool-schemas.1.expected.json` 比对，root 则继续使用默认 schema pin。生成的工具目录会另外铸造一个 child 作用域，以收录同一个作用域局部 schema。
|
||||
手写快照会启动一个可继续 child，执行真实的作用域局部 `report` 工具，观察默认唤醒投递所产生的那一个普通 parent 轮次，然后提交一条后续 parent 提示词，使其消费封装后的报告。它声明 child pin `1`，因此本不属于全局的 `report` schema 与该 child 自身的提示词会分别与 `tool-schemas.1.expected.json` 和 `system-prompt.1.expected.md` 比对，root 则继续使用类别 pin。生成的工具目录会另外铸造一个 child 作用域，以收录同一个作用域局部 schema。
|
||||
|
||||
## 曾考虑的替代方案
|
||||
|
||||
@@ -72,7 +72,7 @@ ACP（Agent Client Protocol）快照 harness 新增 `waitForSubagentTurnEnd`，
|
||||
|
||||
### 始终唤醒 parent
|
||||
|
||||
每次报告都唤醒 parent 会产生未经请求的轮次，还可能沿嵌套 subagent 级联扩散。静默投递更适合作为后台协调的默认值，而需要立即处理的部署可选择 wakeup。
|
||||
每次报告都唤醒 parent 会产生未经请求的轮次，还可能沿嵌套 subagent 级联扩散。当初选择静默投递作为默认值，前提是 parent 还有别的理由去读自己的上下文。[报告义务](2026-08-06-continuable-child-report-obligation.md)取代了该选择：已经停驻的后台协调者并没有这样的理由，因此唤醒成为默认值，而本段现在记录的是 `quiet` 为何仍然保留。
|
||||
|
||||
### 允许 child 选择投递模式
|
||||
|
||||
@@ -103,16 +103,16 @@ ACP（Agent Client Protocol）快照 harness 新增 `waitForSubagentTurnEnd`，
|
||||
- 只有安装 report 包贡献时，可继续进程内 child 才会恰好暴露一个作用域局部 `report` schema；无关 Agent 永远不会暴露该 schema。
|
||||
- 工具返回 parent 消息的稳定 `MessageId`。静默投递没有 `InboxItemId`；唤醒投递会产生一个单独的 inbox 条目实例。
|
||||
- 只有确切的驻留 child 才能报告，且只能报告给根据持久化谱系推导的确切在线直接 parent。服务不接受接收方参数，也不提供离线 fallback。
|
||||
- 静默投递是校验后的默认模式，绝不会启动 parent 请求。wakeup 会恰好创建一个后续 FIFO 轮次，绝不 steering 已开始的轮次。
|
||||
- 唤醒投递是校验后的默认模式：它会恰好创建一个后续 FIFO 轮次，绝不 steering 已开始的轮次。静默投递则绝不会启动 parent 请求。
|
||||
- parent 接受后取消或 dispose child 不会撤回报告。接受前，child dispose、drain、parent 丢失或调用方取消都会拒绝操作。
|
||||
- 新建和恢复的 Activation 都会在发布前组合当前设置贡献。新授权等待下一个 Activation 才生效，而已驻留 child 的授权撤销立即生效。
|
||||
- 单元覆盖固定可见性、allow-list 行为、两种投递模式、稳定的消息与发送方身份、嵌套路由、无效发送方、缺失的 parent、取消、drain、撤销竞争，以及不存在 Task 或隐式最终报告。
|
||||
- 无密钥整体组装快照证明真实 child 工具、静默且不唤醒的行为、持久化 parent 封装，以及 parent 后续消费。
|
||||
- 无密钥整体组装快照证明真实 child 工具、那一个被唤醒的 parent 轮次、持久化 parent 封装，以及 parent 后续消费。
|
||||
|
||||
### 已接受的风险
|
||||
|
||||
该接受边界弱于持久化端到端投递。崩溃可能导致结果不明，重试则可能重复报告。
|
||||
|
||||
wakeup 模式可能在嵌套 child 频繁报告时放大模型工作量。由部署所有者控制并默认静默，可以限制该风险，但无法完全消除。
|
||||
唤醒投递可能在嵌套 child 频繁报告时放大模型工作量。通过 `reportDelivery` 交由部署所有者控制，可以限制该风险，但无法完全消除。
|
||||
|
||||
注册表中的存在性就是 parent 在线信号。宿主拥有的 parent 如果已开始 `AgentHandle.dispose()` 但尚未完成其作用域清理，仍可能接受并追加一条本进程不会再处理的报告。要弥合这个缺口，需要 Agent 层面的 dispose 开始信号，不能由 subagent 层推断。
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-01-windows-pwsh-default.md
|
||||
2026-08-01-windows-pwsh-default.md: f0da86e52bcdd53a10b60164d7cc12261cfc5c49
|
||||
2026-08-01-windows-pwsh-default.zh.md: 41a6429eab8f86a8960ac4aa372aeacfda4661c4
|
||||
2026-08-01-windows-pwsh-default.md: 4e681b32088954d870df86898e26fe2cae669f14
|
||||
2026-08-01-windows-pwsh-default.zh.md: a9d600f8a8e47db49c3733f33091e667e341c6a7
|
||||
|
||||
@@ -12,9 +12,9 @@ The harness's shipped execution profile is bash-first on every platform. Windows
|
||||
|
||||
Windows hosts booting a shipped profile (`dsh web`, `dsh --profile headless`, one-shot tasks) get the PowerShell stack by default; POSIX hosts are unchanged.
|
||||
|
||||
- **The platform layer is a data file, not a roster rewrite.** `@deepseek-ai/dsh-base` ships [`windows.cordis.patch.yml`](../../../../packages/bundle/base/windows.cordis.patch.yml) alongside its universal `cordis.patch.yml`: it disables `bash-sandbox`/`tool-bash` (the POSIX-only executor and its dialect tool) and inserts `pwsh-local`/`tool-pwsh`. Windows has no OS sandbox runner (landlock/bwrap/seatbelt are POSIX-only), so the layer drops the sandbox stack entirely — `sandbox`, `sandbox-policy`, and `fs-sandbox` are disabled and the unconfined `dsh-fs-local` provides `ctx.fs` — and degrades to danger-full-access: `permission`/`ui-permission` leave the roster (dsh-permission requires a confining executor — presets bundle a sandbox mode the unconfined executor cannot honor; see its constructor guard — and the client knob would advertise a boundary that does not exist), and the `approval` service is disabled — nothing in the Windows roster asks for approval, so the model is never told approval exists or that asks are auto-rejected. Keeping fs-only path rules would be theater: the unconfined shell can bypass them with one command, so the honest Windows posture is full access rather than a boundary only the fs tools pretend to enforce.
|
||||
- **The launcher injects the layer by platform.** `apps/cli/src/windows-shell.ts` resolves it from the base bundle layer's `packageDir` between the bundle layers and the user layers on `win32` hosts, in every composition path (boot, config-only HMR recomposition, config dumps). Overriding the shipped default is a composition decision: a Windows host that prefers the bash stack — or confinement — re-enables the bash rows through its profile or home `cordis.patch.yml`. Custom profiles without the base bundle are skipped (they own their shell stack); a base bundle that ships no Windows shell patch fails loud.
|
||||
- **Module resolution is restored for cold starts.** The profiles-rework CLI dropped the pwsh packages from `apps/cli`'s dependency closure, so `healProfilesModuleFallback` never linked them into `$DSH_HOME/profiles/node_modules` and a fresh Windows host could not resolve the inserted rows. `apps/cli` and `dsh-base` re-declare `dsh-pwsh-local`/`dsh-tool-pwsh`, and `dsh-base` also declares `dsh-fs-local`; the base bundle lists every row plugin as a dependency by house style.
|
||||
- **The platform layer is a data file, not a roster rewrite.** `@deepseek-ai/dsh-base` ships [`windows.cordis.patch.yml`](../../../../packages/bundle/base/windows.cordis.patch.yml) alongside its universal `cordis.patch.yml`. It disables the POSIX-only `bash-sandbox`/`tool-bash` rows and inserts `pwsh-sandbox`/`tool-pwsh`. The later [Windows ACL sandbox decision](2026-08-08-windows-acl-restricted-token-sandbox.md) filled the win32 runner chain and superseded this note's original unconfined roster: `sandbox`, `sandbox-policy`, `fs-sandbox`, `permission`/`ui-permission`, and `approval` now stay enabled exactly as on POSIX, while the ACL backend truthfully reports its Everyone and hard-link gaps as partial enforcement.
|
||||
- **The launcher injects the layer by platform.** `apps/cli/src/windows-shell.ts` resolves it from the base bundle layer's `packageDir` between the bundle layers and the user layers on `win32` hosts, in every composition path (boot, config-only HMR recomposition, config dumps). Overriding the shipped default is a composition decision: a Windows host that prefers the bash stack re-enables the bash rows and disables both pwsh rows through its profile or home `cordis.patch.yml`. Custom profiles without the base bundle are skipped (they own their shell stack); a base bundle that ships no Windows shell patch fails loud.
|
||||
- **Module resolution is restored for cold starts.** The profiles-rework CLI dropped the pwsh packages from `apps/cli`'s dependency closure, so `healProfilesModuleFallback` never linked them into `$DSH_HOME/profiles/node_modules` and a fresh Windows host could not resolve the inserted rows. `apps/cli` and `dsh-base` declare `dsh-pwsh-sandbox`/`dsh-tool-pwsh`; the executor's dependency chain supplies `dsh-pwsh-local`, and the base bundle lists every row plugin as a dependency by house style.
|
||||
|
||||
The pwsh GUI rendering shipped earlier with the [pwsh UI presentation matches bash decision](2026-08-05-pwsh-ui-bash-parity.md); the [pwsh tool bash parity decision](2026-08-02-pwsh-tool-bash-parity.md) ships the tool's surface. Nothing in this decision changes POSIX behavior.
|
||||
|
||||
@@ -24,21 +24,21 @@ The pwsh GUI rendering shipped earlier with the [pwsh UI presentation matches ba
|
||||
|
||||
**Ship the platform layer from `apps/cli` code instead of a bundle data file.** Rejected: the patch belongs next to the rows it replaces, in the bundle that owns them, so the shipped roster stays visible as composition data and dumps carry its provenance; the launcher contributes only the win32 gate.
|
||||
|
||||
**Keep `permission`/`ui-permission` on Windows.** Rejected: `dsh-permission` hard-requires `ctx.bash.sandboxMode` and fails loud at load over an unconfined executor; making it tolerate an unconfined shell would advertise presets the shell cannot honor.
|
||||
**Keep `permission`/`ui-permission` on Windows without a confining runner.** Rejected by the original delivery: `dsh-permission` hard-requires `ctx.bash.sandboxMode` and fails loud at load over an unconfined executor. The later ACL runner removed that premise, so the current roster retains both rows.
|
||||
|
||||
**Keep fs path-rule confinement on Windows (`sandbox-policy` + `fs-sandbox` without OS runners).** Rejected: the shell is the model's primary tool and unconfined on Windows, so fs-only path rules are trivially bypassable and would overstate the boundary; the honest posture is full degradation to danger-full-access.
|
||||
**Keep fs path-rule confinement on Windows without an OS runner.** Rejected by the original delivery: an unconfined shell could bypass fs-only path rules. The current ACL runner confines the shell and the fs provider under one policy, so this rejected half-boundary is no longer the shipped shape.
|
||||
|
||||
**Ship a `DSH_WINDOWS_SHELL` environment escape hatch.** Rejected: decisive behavior changes belong in composition config, which already overrides the platform layer row by id; a second override channel would split the single source of truth for roster decisions.
|
||||
|
||||
## Consequences
|
||||
|
||||
- A Windows host running a shipped `dsh` surface gets `pwsh` as its shell tool and PowerShell as the `ctx.bash` executor without configuration; `bash` is absent from the model-visible roster there (its tool row is disabled).
|
||||
- Windows has no sandbox at all: the fs tools run unconfined (`dsh-fs-local`), the approval service is absent (nothing asks for approval, and the model is never told approval exists), and the permission switcher is gone. The model-visible posture is honest full access rather than a boundary the shell can bypass.
|
||||
- Windows commands and fs operations share the sandbox policy, permission switcher, and approval service. The ACL runner confines writes but reports `enforcement: 'partial'`; explicit `danger-full-access` remains the approved bypass rather than the platform default.
|
||||
- POSIX hosts are unchanged: the platform layer never applies, and the bash stack remains the universal `cordis.patch.yml` rows.
|
||||
- Windows hosts that prefer the bash stack (e.g. with WSL/Git-Bash on PATH) override the shipped default through their profile or home `cordis.patch.yml` — disabling `pwsh-local`/`tool-pwsh` and re-enabling `bash-sandbox`/`tool-bash` (both executors register the same `bash` service, so an incomplete recipe fails loud at load) — composition config is the one override channel.
|
||||
- Windows hosts that prefer the bash stack (e.g. with WSL/Git-Bash on PATH) override the shipped default through their profile or home `cordis.patch.yml` — disabling `pwsh-sandbox`/`tool-pwsh` and re-enabling `bash-sandbox`/`tool-bash` (both executors register the same `bash` service, so an incomplete recipe fails loud at load) — composition config is the one override channel.
|
||||
|
||||
## Verification
|
||||
|
||||
- Unit: `apps/cli/tests/windows-shell.spec.ts` pins the win32 default, the custom-profile skip, and the missing-patch failure with the platform injected, and composes the REAL shipped bundle layers (dsh-base + dsh-web-app resolved from the app installation) through the boot's patch algorithm to assert the win32 danger-full-access roster and the base-only-profile warning; `packages/bundle/base/tests/base.spec.ts` pins the shipped Windows patch file shape (disables, inserts, and the absent approval service).
|
||||
- Unit: `apps/cli/tests/windows-shell.spec.ts` pins the win32 default, custom-profile skip, missing-patch failure, cold-start dependency closure, and real composed roster; `packages/bundle/base/tests/base.spec.ts` pins that the Windows layer disables only the bash rows, inserts the confined pwsh rows, and leaves sandbox, permission, fs, and approval ownership untouched.
|
||||
- Keyless: a win32 `dsh --profile <name> --dump-config` shows the pwsh rows with `windows.cordis.patch.yml` provenance and the bash rows disabled; the POSIX dump (CI Linux) is unchanged.
|
||||
- The real-composition smoke boots the web profile on win32 with the pwsh stack mounted (the exact roster this note describes).
|
||||
|
||||
@@ -12,9 +12,9 @@ harness 交付的执行画像在每个平台都是 bash 优先。Windows 主机
|
||||
|
||||
启动交付 profile（`dsh web`、`dsh --profile headless`、一次性任务）的 Windows 主机默认获得 PowerShell 栈；POSIX 主机不变。
|
||||
|
||||
- **平台层是数据文件，不是清单重写。** `@deepseek-ai/dsh-base` 随通用 `cordis.patch.yml` 一起交付 [`windows.cordis.patch.yml`](../../../../packages/bundle/base/windows.cordis.patch.yml)：它禁用 `bash-sandbox`/`tool-bash`（仅 POSIX 的执行器及其方言工具）并插入 `pwsh-local`/`tool-pwsh`。Windows 上没有 OS 级 sandbox runner（landlock/bwrap/seatbelt 均为 POSIX 专属），因此该层整体移除 sandbox 栈——`sandbox`、`sandbox-policy`、`fs-sandbox` 被禁用，由不限权的 `dsh-fs-local` 提供 `ctx.fs`——并完全退化为 danger-full-access：`permission`/`ui-permission` 离开清单（dsh-permission 要求有限权能力的执行器——preset 捆绑的是无限制执行器无法兑现的 sandbox 模式；见其构造函数守卫——客户端旋钮会宣传一个并不存在的边界），`approval` 服务也被禁用——Windows 清单里没有任何动作需要审批，模型也不会被告知"审批存在"或"请求会被自动拒绝"。保留仅限 fs 的路径规则是摆设：不限权的 shell 一条命令即可绕过，因此诚实的 Windows 姿态是全权访问，而不是一个只有 fs 工具假装执行的边界。
|
||||
- **启动器按平台注入该层。** `apps/cli/src/windows-shell.ts` 在 `win32` 主机上从 base bundle 层的 `packageDir` 解析它，置于 bundle 层与用户层之间，覆盖所有组合路径（启动、config-only HMR 重组合、配置转储）。覆盖交付默认是组合决策：偏好 bash 栈（或偏好有限权）的 Windows 主机通过其 profile 或 home 的 `cordis.patch.yml` 重新启用 bash 行。未挂 base bundle 的自定义 profile 被跳过（它们自己拥有 shell 栈）；base bundle 缺 `windows.cordis.patch.yml` 时 fail loud。
|
||||
- **冷启动的模块解析已恢复。** profiles 重构把 pwsh 包从 `apps/cli` 的依赖闭包中删掉了，`healProfilesModuleFallback` 因此从未把它们链接进 `$DSH_HOME/profiles/node_modules`，新 Windows 主机解析不到插入的行。`apps/cli` 与 `dsh-base` 重新声明 `dsh-pwsh-local`/`dsh-tool-pwsh`，`dsh-base` 还声明 `dsh-fs-local`；按仓库惯例，base bundle 把每个行插件都列为依赖。
|
||||
- **平台层是数据文件，不是清单重写。** `@deepseek-ai/dsh-base` 随通用 `cordis.patch.yml` 一起交付 [`windows.cordis.patch.yml`](../../../../packages/bundle/base/windows.cordis.patch.yml)。它禁用仅限 POSIX 的 `bash-sandbox`/`tool-bash` 行，并插入 `pwsh-sandbox`/`tool-pwsh`。后续的 [Windows ACL 沙箱决策](2026-08-08-windows-acl-restricted-token-sandbox.md)填充了 win32 runner 链，并取代了本笔记最初的不限权清单：`sandbox`、`sandbox-policy`、`fs-sandbox`、`permission`/`ui-permission` 与 `approval` 均与 POSIX 上一样保持启用，而 ACL 后端则如实把 Everyone 与硬链接缺口报告为部分强制执行。
|
||||
- **启动器按平台注入该层。** `apps/cli/src/windows-shell.ts` 在 `win32` 主机上从 base bundle 层的 `packageDir` 解析它，置于 bundle 层与用户层之间，覆盖所有组合路径（启动、config-only HMR 重组合、配置转储）。覆盖交付默认是组合决策：偏好 bash 栈的 Windows 主机通过其 profile 或 home 的 `cordis.patch.yml` 重新启用 bash 行，并禁用两个 pwsh 行。未挂 base bundle 的自定义 profile 被跳过（它们自己拥有 shell 栈）；base bundle 缺 `windows.cordis.patch.yml` 时 fail loud。
|
||||
- **冷启动的模块解析已恢复。** profiles 重构把 pwsh 包从 `apps/cli` 的依赖闭包中删掉了，`healProfilesModuleFallback` 因此从未把它们链接进 `$DSH_HOME/profiles/node_modules`，新 Windows 主机解析不到插入的行。`apps/cli` 与 `dsh-base` 声明 `dsh-pwsh-sandbox`/`dsh-tool-pwsh`；执行器的依赖链提供 `dsh-pwsh-local`，按仓库惯例，base bundle 把每个行插件都列为依赖。
|
||||
|
||||
pwsh GUI 渲染已随 [pwsh UI 呈现与 bash 对齐决策](2026-08-05-pwsh-ui-bash-parity.md) 先行交付；[pwsh 工具与 bash 对齐决策](2026-08-02-pwsh-tool-bash-parity.md) 交付了工具表面。本决策不改变任何 POSIX 行为。
|
||||
|
||||
@@ -24,21 +24,21 @@ pwsh GUI 渲染已随 [pwsh UI 呈现与 bash 对齐决策](2026-08-05-pwsh-ui-b
|
||||
|
||||
**从 `apps/cli` 代码而非 bundle 数据文件交付平台层。** 否决：patch 应放在它替换的行旁边、属于拥有这些行的 bundle，让交付清单作为组合数据保持可见、转储带有出处；启动器只贡献 win32 门控。
|
||||
|
||||
**在 Windows 上保留 `permission`/`ui-permission`。** 否决：`dsh-permission` 硬性要求 `ctx.bash.sandboxMode`，在无限制执行器上加载即 fail loud；让它容忍无限制 shell 会宣传 shell 无法兑现的 preset。
|
||||
**在 Windows 没有隔离 runner 时保留 `permission`/`ui-permission`。** 最初交付时否决：`dsh-permission` 硬性要求 `ctx.bash.sandboxMode`，并在不限权执行器上加载时 fail loud。后续的 ACL runner 消除了该前提，因此当前清单保留这两行。
|
||||
|
||||
**在 Windows 上保留 fs 路径规则限制（无 OS runner 的 `sandbox-policy` + `fs-sandbox`）。** 否决：shell 是模型的主工具且在 Windows 上不限权，仅限 fs 的路径规则一行命令即可绕过，会夸大边界；诚实的姿态是完全退化到 danger-full-access。
|
||||
**在 Windows 没有 OS runner 时保留 fs 路径规则限制。** 最初交付时否决：不限权 shell 可以绕过仅限 fs 的路径规则。当前 ACL runner 用同一策略约束 shell 与 fs 提供方，因此这项被否决的半边界已不是当前交付形态。
|
||||
|
||||
**交付 `DSH_WINDOWS_SHELL` 环境变量逃生门。** 否决：决定性的行为变更应集中在组合配置中，而组合配置已能按行 id 覆盖平台层；第二条覆盖通道会分裂清单决策的单一事实来源。
|
||||
|
||||
## 后果
|
||||
|
||||
- 运行交付版 `dsh` 表面的 Windows 主机无需配置即获得 `pwsh` 作为 shell 工具、PowerShell 作为 `ctx.bash` 执行器；那里的模型可见清单中没有 `bash`（其工具行被禁用）。
|
||||
- Windows 上没有任何沙箱：fs 工具不限权运行（`dsh-fs-local`）、`approval` 服务不存在（没有任何动作需要审批，模型也不会被告知审批存在）、权限切换器消失。模型可见的姿态是诚实的全权访问，而不是一个 shell 可以绕过的边界。
|
||||
- Windows 命令与 fs 操作共用沙箱策略、权限切换器和 approval 服务。ACL runner 限制写入，但报告 `enforcement: 'partial'`；显式的 `danger-full-access` 仍是获准的绕过方式，而非平台默认。
|
||||
- POSIX 主机不变：平台层永不生效，bash 栈仍是通用 `cordis.patch.yml` 的行。
|
||||
- 偏好 bash 栈的 Windows 主机（例如 PATH 上有 WSL/Git-Bash 时）通过其 profile 或 home 的 `cordis.patch.yml` 覆盖交付默认——禁用 `pwsh-local`/`tool-pwsh` 并重新启用 `bash-sandbox`/`tool-bash`（两个执行器注册同一个 `bash` 服务，配方不完整会在加载时 fail loud）——组合配置是唯一的覆盖通道。
|
||||
- 偏好 bash 栈的 Windows 主机（例如 PATH 上有 WSL/Git-Bash 时）通过其 profile 或 home 的 `cordis.patch.yml` 覆盖交付默认——禁用 `pwsh-sandbox`/`tool-pwsh` 并重新启用 `bash-sandbox`/`tool-bash`（两个执行器注册同一个 `bash` 服务，配方不完整会在加载时 fail loud）——组合配置是唯一的覆盖通道。
|
||||
|
||||
## 验证
|
||||
|
||||
- 单元：`apps/cli/tests/windows-shell.spec.ts` 以平台注入固定 win32 默认、自定义 profile 跳过与缺文件失败，并通过启动所用的 patch 算法组合真实交付的 bundle 层（从应用安装解析的 dsh-base + dsh-web-app）断言 win32 danger-full-access 清单与 base-only profile 警告；`packages/bundle/base/tests/base.spec.ts` 固定交付的 Windows patch 文件形状（禁用、插入与缺席的 approval 服务）。
|
||||
- 单元：`apps/cli/tests/windows-shell.spec.ts` 固定 win32 默认、自定义 profile 跳过、缺少 patch 时失败、冷启动依赖闭包和真实组合清单；`packages/bundle/base/tests/base.spec.ts` 固定 Windows 层仅禁用 bash 行、插入受限的 pwsh 行，并且不改变沙箱、权限、fs 与审批的归属。
|
||||
- Keyless：win32 上的 `dsh --profile <name> --dump-config` 显示带 `windows.cordis.patch.yml` 出处的 pwsh 行、被禁用的 bash 行；POSIX 转储（CI Linux）不变。
|
||||
- 真实组合冒烟在 win32 上启动 web profile，pwsh 栈挂载成功（即本笔记描述的确切清单）。
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-03-fs-tool-error-remedy.md
|
||||
2026-08-03-fs-tool-error-remedy.md: c4f0bb86100a4856c536edf22e520662e19ea4f6
|
||||
2026-08-03-fs-tool-error-remedy.md: 1829a47363079aacc2090e166e94b5c524c87deb
|
||||
2026-08-03-fs-tool-error-remedy.zh.md: 88309bd630732794dfb2401b60ad799a3d63146b
|
||||
|
||||
@@ -21,7 +21,7 @@ In `edit.ts` the `fs/edit-intent` waterfall now sits inside the same `try` as th
|
||||
|
||||
## Alternatives considered
|
||||
|
||||
- **Append the remedy to the provider messages in `dsh-fs` / `dsh-fs-local`.** Rejected because those messages are machine-oriented seam vocabulary consumed by retry, permission, and UI layers as well as the model surface; model-facing wording belongs at the model boundary, where `dsh-tool-fs` already owns result formatting ([filesystem capability seam](../architecture/2026-06-17-filesystem-capability-seam.md)).
|
||||
- **Append the remedy to the provider messages in `dsh-fs` / `dsh-fs-local`.** Rejected because those messages are machine-oriented seam vocabulary consumed by retry, permission, UI, and model-facing layers; model-facing wording belongs at the model boundary, where `dsh-tool-fs` already owns result formatting ([filesystem capability seam](../architecture/2026-06-17-filesystem-capability-seam.md)).
|
||||
- **Add the recovery to prompt guidance instead.** Rejected because the failure arrives mid-task; a static instruction does not reliably reach the retry decision, while the error message is present exactly when the model must act.
|
||||
- **Signal the remedy with a new `FsError` code.** Rejected because the two failures are the same conditions retry layers already handle; splitting the code would fork routing on identical semantics.
|
||||
|
||||
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-05-per-agent-tool-presentation.md
|
||||
2026-08-05-per-agent-tool-presentation.md: 348f7ab0a26e9b39057dbac885304e0d52e0b1fb
|
||||
2026-08-05-per-agent-tool-presentation.zh.md: 4920ee6eb061d44934bfc9f5176e244f5aac8553
|
||||
2026-08-05-per-agent-tool-presentation.md: adb93b51c73d341c153b8fcafe2a08f0a5598478
|
||||
2026-08-05-per-agent-tool-presentation.zh.md: fa83bd4daec9d8e6e9e43295bb81af97782b1fdf
|
||||
|
||||
@@ -12,16 +12,16 @@ The naive reading of "move tools down to the agent plane" does not work. `ctx.to
|
||||
|
||||
## Decision
|
||||
|
||||
Split the registry from its projection. The registry stays host-plane; the **presentation** becomes per-agent state inside it, alongside the per-agent restrictions and guards that already live there.
|
||||
Split the registry from its projection. The registry stays host-plane; the **presentation** becomes scope state inside it, alongside the scoped restrictions and guards that already live there.
|
||||
|
||||
`ToolRegistry.presentAs(mode)` is scoped-only and mirrors `restrict()`: it writes one cell on the calling scope's `ToolLayer` through `ScopedLayers.effect`, so it unwinds with the agent that declared it. `modeFor(scope)` resolves that cell against the config `mode`, which becomes the default for agents declaring nothing rather than a process-wide fact. The three reads that decided presentation — the wire schemas, the `run_code` entry in the visibility view, and the generated SDK section — take the scope's mode instead of the service's.
|
||||
`ToolRegistry.presentAs(mode)` is scoped-only and mirrors `restrict()`: it writes one cell on the calling scope's `ToolLayer` through `ScopedLayers.effect`, so it unwinds with the scope that declared it. In the shipped Web surface that scope is an agent preset's standing mount — the `code` preset carries the `tool-mode` row — so one declaration covers every agent joined to that preset, and `modeFor(scope)` takes the nearest declaration on the chain. It resolves against the config `mode`, which becomes the default for scopes declaring nothing rather than a process-wide fact. The three reads that decided presentation — the wire schemas, the `run_code` entry in the visibility view, and the generated SDK section — take the scope's mode instead of the service's.
|
||||
|
||||
Two consequences fell out and are load-bearing:
|
||||
|
||||
- **`run_code` is appended per scope.** Previously the transport entered every view whenever the transport existed. Per-agent, a native agent must not find `run_code` in its dispatch table because some other agent in the process presents it — so the append is conditional on that scope's own mode, and the transport is built lazily on first need.
|
||||
- **The reserved name is now unconditional.** `run_code` was rejected as a registration only while a code mode was configured. Any agent may now select a code mode, so a name that was free to take under a native deployment would become a collision the moment a preset mounted.
|
||||
|
||||
The SDK prompt section is registered globally by a code-mode deployment (unchanged) and additionally per agent by `presentAs`, where it shadows by name. Its body renders empty for a native scope, which the prompt renderer drops — that is what keeps an agent opting OUT of a code-mode deployment free of an SDK section.
|
||||
The SDK prompt section is registered globally by a code-mode deployment (unchanged) and additionally per scope by `presentAs`, where it shadows by name. Its body renders empty for a native scope, which the prompt renderer drops — that is what keeps an agent opting OUT of a code-mode deployment free of an SDK section.
|
||||
|
||||
The preset expresses the choice through one row, `@deepseek-ai/dsh-agent-tool-mode`, whose whole body is a `presentAs` call. A code mode waits for `ctx.codeRuntime` through `ctx.inject` rather than assuming it: the runtime is host-plane, and a pending row is what `dsh-agent-presets` already reports as an unusable mount, naming the row — so a preset selecting Code Mode against a runtime-less deployment fails where an operator can act.
|
||||
|
||||
|
||||
@@ -12,16 +12,16 @@ agent preset 已经能按会话组装一个 agent 的工具，却管不了这些
|
||||
|
||||
## Decision
|
||||
|
||||
把注册表和它的投影拆开。注册表留在宿主平面；**呈现方式**变成它内部按 agent 的状态，与已经住在那里的按 agent 限制和守卫并列。
|
||||
把注册表和它的投影拆开。注册表留在宿主平面；**呈现方式**变成它内部按 scope 的状态，与已经住在那里的作用域限制和守卫并列。
|
||||
|
||||
`ToolRegistry.presentAs(mode)` 只接受 scoped 上下文，形状照抄 `restrict()`：它通过 `ScopedLayers.effect` 在调用方 scope 的 `ToolLayer` 上写一个单元，因此会随声明它的那个 agent 一起卸载。`modeFor(scope)` 将该单元与 config 的 `mode` 一并解析，后者于是成为「未作声明的 agent」的默认值，而不再是进程级事实。原先决定呈现方式的三处读取——wire schema、可见性视图里的 `run_code` 条目、以及生成的 SDK 段——改为读取该 scope 的模式，而非服务的。
|
||||
`ToolRegistry.presentAs(mode)` 只接受 scoped 上下文，形状照抄 `restrict()`：它通过 `ScopedLayers.effect` 在调用方 scope 的 `ToolLayer` 上写一个单元，因此会随声明它的那个 scope 一起卸载。在随附的 Web 界面里那个 scope 是某个 agent preset 的常驻挂载——`code` preset 携带 `tool-mode` 行——因此一份声明覆盖加入该 preset 的每个 agent，而 `modeFor(scope)` 取作用域链上最近的那份声明。它与 config 的 `mode` 一并解析，后者于是成为「未作声明的 scope」的默认值，而不再是进程级事实。原先决定呈现方式的三处读取——wire schema、可见性视图里的 `run_code` 条目、以及生成的 SDK 段——改为读取该 scope 的模式，而非服务的。
|
||||
|
||||
有两个随之而来的结果，且都是承重的：
|
||||
|
||||
- **`run_code` 按 scope 追加。** 此前只要传输存在，它就进入每一个视图。按 agent 之后，一个 native agent 不能因为进程里别的 agent 呈现了它、就在自己的分发表里看到 `run_code`——因此这次追加以该 scope 自身的模式为条件，传输也改为首次需要时才构建。
|
||||
- **保留名现在无条件生效。** `run_code` 此前只在配置了 code 模式时才被拒绝注册。如今任何 agent 都可能选择 code 模式，因此一个在 native 部署下可以随便占用的名字，会在某个 preset 挂载的那一刻变成冲突。
|
||||
|
||||
SDK 提示词段由 code 模式的部署全局注册（不变），并由 `presentAs` 额外按 agent 注册一份，后者按名字遮蔽前者。它的正文对 native scope 渲染为空，而提示词渲染器会丢弃空段——正是这一点让「在 code 模式部署下选择退出」的 agent 不带 SDK 段。
|
||||
SDK 提示词段由 code 模式的部署全局注册（不变），并由 `presentAs` 额外按 scope 注册一份，后者按名字遮蔽前者。它的正文对 native scope 渲染为空，而提示词渲染器会丢弃空段——正是这一点让「在 code 模式部署下选择退出」的 agent 不带 SDK 段。
|
||||
|
||||
preset 用一行来表达这个选择：`@deepseek-ai/dsh-agent-tool-mode`，其全部内容就是一次 `presentAs` 调用。code 类模式通过 `ctx.inject` 等待 `ctx.codeRuntime` 而非假定它存在：运行时在宿主平面，而一个 pending 的行正是 `dsh-agent-presets` 已经会报告的「不可用挂载」并会指名该行——于是在无运行时的部署上选择 Code Mode 的 preset，会在操作者能够动手的地方失败。
|
||||
|
||||
|
||||
@@ -0,0 +1,6 @@
|
||||
# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-06-continuable-child-report-obligation.md
|
||||
2026-08-06-continuable-child-report-obligation.md: f152ec1b8c353f094f2ba70785112eb1e165c510
|
||||
2026-08-06-continuable-child-report-obligation.zh.md: 4ec17e4642ffac385e6ce5464f41a3f4b3bebdbf
|
||||
@@ -0,0 +1,60 @@
|
||||
# Agent Note: The continuable child return channel is an obligation
|
||||
|
||||
Status: implemented
|
||||
|
||||
English | [中文](2026-08-06-continuable-child-report-obligation.zh.md)
|
||||
|
||||
## Problem
|
||||
|
||||
A continuable background child owns its own Session, so nothing it writes there reaches the agent that started it. [The report tool](2026-07-30-continuable-subagent-report-tool.md) gave that child a return channel and then presented it as one option among several: the schema said "call this zero or more times", nothing in the child's prompt asked it to call the tool at all, and the accepted default scheduling (`quiet`) added the report to a parked parent's next request without waking it.
|
||||
|
||||
Each of those choices is defensible alone. Together they made the return channel unusable as a delegation contract. A child that finished its work, wrote its answer into its own transcript, and stopped left the parent with nothing; a child that did report reached a parent that had already parked and would not read the report until something unrelated woke it. External reports of parents busy-polling `list_agents`, re-sending messages to settled children, and abandoning `subagent` for `workflow` all reduce to the same missing guarantee.
|
||||
|
||||
## Decision
|
||||
|
||||
The return channel is an instruction the child receives, not a capability it may discover. The report package installs two scope-local registrations into every continuable in-process child, and one disposer revokes both:
|
||||
|
||||
- the `report` tool, whose description now states that the child calls it once before finishing with a self-contained final result, and earlier for progress that changes what the parent should do next;
|
||||
- a `tool:report` system-prompt section at order 117 carrying the same obligation in the child's own voice, so a child that never reads tool descriptions closely still receives it.
|
||||
|
||||
`reportDelivery` now defaults to `wakeup`. An accepted report creates exactly one ordinary later parent turn and wakes a parked parent driver; it still never steers an open turn. `quiet` remains available for deployments that prefer unread reports over turn amplification.
|
||||
|
||||
### Why the section and the description both exist
|
||||
|
||||
They address different failure modes. The tool description is read when the model is already considering `report`; the prompt section is read when it is deciding whether it is finished. The obligation belongs at both points because the failure this fixes — a child that simply stops — happens at the second one.
|
||||
|
||||
The section is registered on the child's own scope, the same mechanism [child composition](../../../../packages/subagent/subagent/src/child-agent.ts) already uses for a shadowing persona, so the parent and every sibling see neither the tool nor the guidance. `installReportTool` rolls the section back if tool registration fails, and its returned disposer attempts both revocations before surfacing cleanup failures.
|
||||
|
||||
### Instruction, not enforcement
|
||||
|
||||
Nothing rejects a child that never reports. No runtime path inspects whether a report was sent, and `report` still accepts zero or many calls per turn. The change is model-facing wording plus a scheduling default; the service authority, acknowledgement, and recovery contracts are unchanged.
|
||||
|
||||
That boundary is deliberate: prompt text can only reach a child that is still running its own loop. A child stopped by an error, a token ceiling, cancellation, or teardown never gets the chance to comply, which is why the runtime keeps its own account of settlement rather than trusting this instruction ([manager-owned settlement delivery](2026-08-06-manager-owned-subagent-settlement-delivery.md)).
|
||||
|
||||
### Snapshot coverage
|
||||
|
||||
The assembled ACP `subagent-report` scenario now exercises the shipped default: the child reports, the parked parent takes one ordinary turn on that report, and a later prompt still reads the report back out of the durable log. Because the child's scope now composes a prompt the class pin cannot describe, the snapshot harness gained `pinsChildSystemPrompts`, the exact counterpart of the existing `pinsChildToolSchemas`: it moves one child fixture's prompt into `system-prompt.<n>.expected.md`, leaves every other request-header field to the class pin, requires the sidecar exactly when declared, and rejects a sidecar identical to that class pin so a redundant copy cannot drift.
|
||||
|
||||
## Alternatives considered
|
||||
|
||||
**Keep `quiet` as the default and rely on the prompt alone.** This was the shipped position, and it supersedes nothing on its own: a report the parent never reads is indistinguishable from a report never sent. The [report-tool note's](2026-07-30-continuable-subagent-report-tool.md) rejection of always-waking assumed the parent had another reason to look at its context; a parked background coordinator does not. Turn amplification is the real cost, and it is now the reason `quiet` still exists rather than the reason it is the default.
|
||||
|
||||
**Let the child choose the delivery mode per call.** Unchanged from the original rejection: the model would own scheduler pressure, and behavior would vary per call rather than per deployment.
|
||||
|
||||
**Put the obligation only in the tool description.** A description is read while choosing among tools. The child this change targets is not choosing a tool; it believes it is done. Prompt guidance is the surface that reaches that decision.
|
||||
|
||||
**Enforce the obligation at settlement by rejecting a silent child.** There is nothing to reject: by the time settlement is observable the child's loop is over, and failing its teardown would destroy work rather than deliver it. Delivering the terminal facts unconditionally from the runtime is the answer to that case, and it belongs to the continuation manager, not to this package.
|
||||
|
||||
## Consequences
|
||||
|
||||
- Every continuable in-process child with this package loaded carries one extra prompt section and a longer `report` description in every request; no other Agent's request changes.
|
||||
- The default deployment wakes the parent once per accepted report. A nested tree that reports frequently consumes extra parent turns; `quiet` is the documented escape.
|
||||
- `installReportTool` requires `ctx.systemPrompt` in the child scope, so the package declares `systemPrompt` in `inject` and fails at load rather than at the next child materialization.
|
||||
- Unit coverage pins the new default, two load-bearing instruction phrases, the section's child-only scope against both the parent and a sibling, and rollback or revocation of both registrations.
|
||||
- Three assembled ACP scenarios with continuable children pin the complete instruction text through the new sidecar; a future change to any child-scoped section fails those scenarios instead of passing silently.
|
||||
|
||||
### Accepted risks
|
||||
|
||||
Waking by default amplifies model work in deep trees. The deployment owns that through `reportDelivery`, and the amplification is bounded by one turn per accepted report.
|
||||
|
||||
A child can still finish without reporting, and this change cannot detect it. Only the runtime's own [settlement account](2026-08-06-manager-owned-subagent-settlement-delivery.md) closes that case.
|
||||
@@ -0,0 +1,60 @@
|
||||
# Agent Note: 可继续 child 的返回通道是一项义务
|
||||
|
||||
Status: implemented
|
||||
|
||||
[English](2026-08-06-continuable-child-report-obligation.md) | 中文
|
||||
|
||||
## 问题
|
||||
|
||||
可继续后台 child 拥有自己的 Session，因此它写在那里的任何内容都不会到达启动它的 agent。[report 工具](2026-07-30-continuable-subagent-report-tool.md)为该 child 提供了一条返回通道，却把它呈现为若干选项之一：schema 里写着「可调用零次或多次」，child 的提示词中没有任何地方要求它调用该工具，而已采纳的默认调度（`quiet`）会把报告加入已停驻 parent 的下一次请求，却不唤醒它。
|
||||
|
||||
这些选择单独看都站得住脚。合在一起，它们让这条返回通道无法作为委派契约使用。一个完成工作、把答案写进自己 transcript（文本记录）随后停止的 child，会让 parent 一无所获；而确实上报了的 child，面对的是一个已经停驻、要等到别的事件把它唤醒才会读到报告的 parent。外部反馈中的 parent 忙轮询 `list_agents`、反复向已结算 child 发送消息、以及放弃 `subagent` 改用 `workflow`，都可归结为同一处缺失的保证。
|
||||
|
||||
## 决策
|
||||
|
||||
返回通道是 child 收到的一条指令，而不是它需要自行发现的能力。report 包会向每个可继续进程内 child 安装两项作用域局部注册，并由同一个 disposer 撤销两者：
|
||||
|
||||
- `report` 工具，其描述现在说明 child 要在结束前调用一次并给出自足的最终结果，并在部分进展会改变 parent 下一步动作时提前调用；
|
||||
- 一个 order 为 117 的 `tool:report` 系统提示词 section，用 child 自己的语气承载同一条义务，使从不细读工具描述的 child 仍能收到它。
|
||||
|
||||
`reportDelivery` 的默认值现在是 `wakeup`。一条被接受的报告恰好创建一个普通的后续 parent 轮次并唤醒停驻的 parent 驱动；它仍然绝不 steering（中途引导）已开始的轮次。对于宁可让报告无人阅读也要避免轮次放大的部署，`quiet` 依旧可用。
|
||||
|
||||
### 为什么 section 与描述同时存在
|
||||
|
||||
两者针对不同的失效模式。工具描述是在模型已经在考虑 `report` 时被读到的；提示词 section 是在它判断自己是否已经完成时被读到的。这条义务必须同时出现在两处，因为本次修复的失效——child 直接停下——发生在第二处。
|
||||
|
||||
该 section 注册在 child 自己的作用域上，与[child 组合](../../../../packages/subagent/subagent/src/child-agent.ts)为遮蔽式 persona 已经使用的机制相同，因此 parent 与所有同级都看不到该工具与该指引。工具注册失败时，`installReportTool` 会回滚该 section；它返回的 disposer 会先尝试撤销两项注册，再抛出清理失败。
|
||||
|
||||
### 是指令，不是强制
|
||||
|
||||
没有任何东西会拒绝一个从不上报的 child。没有任何运行时路径会检查是否发送过报告，`report` 仍接受一个轮次中调用零次或多次。本次改动是面向模型的措辞加上一个调度默认值；服务权限、确认与恢复契约都保持不变。
|
||||
|
||||
这条边界是刻意划定的：提示词文本只能到达仍在运行自身循环的 child。被错误、token 上限、取消或拆卸终止的 child 根本没有机会遵守，因此运行时会自己记录结算这件事，而不是信任这条指令（见[由管理器负责的结算投递](2026-08-06-manager-owned-subagent-settlement-delivery.md)）。
|
||||
|
||||
### 快照覆盖
|
||||
|
||||
整体组装的 ACP `subagent-report` 场景现在演练随附的默认行为：child 上报，停驻的 parent 就该报告执行一个普通轮次，随后的提示词仍能从持久化日志中把报告读回来。由于该 child 的作用域现在组合出类别 pin 无法描述的提示词，快照 harness 新增了 `pinsChildSystemPrompts`，它与既有 `pinsChildToolSchemas` 完全对称：把一个 child fixture 的提示词移入 `system-prompt.<n>.expected.md`，其余请求 header 字段仍归类别 pin 所有，要求 sidecar 恰好在声明时存在，并拒绝与该类别 pin 完全相同的 sidecar，使冗余副本无法悄悄漂移。
|
||||
|
||||
## 备选方案
|
||||
|
||||
**保留 `quiet` 作为默认值，只依赖提示词。** 这曾是随附的立场，而它本身什么也没有解决：一条 parent 从不阅读的报告，与一条从未发送的报告无法区分。[report 工具 Agent Note](2026-07-30-continuable-subagent-report-tool.md)对「始终唤醒」的否决，前提是 parent 还有别的理由去查看自己的上下文；已停驻的后台协调者并没有。轮次放大才是真正的代价，而它现在是 `quiet` 仍然保留的理由，而不是它作为默认值的理由。
|
||||
|
||||
**让 child 按调用选择投递模式。** 与最初的否决相同：模型将掌握调度压力，行为也会随调用而非随部署变化。
|
||||
|
||||
**只把义务写在工具描述里。** 描述是在从多个工具中选择时被读到的。本次改动针对的 child 并不在选择工具，它认为自己已经做完了。提示词指引才是能触及该判断的界面。
|
||||
|
||||
**在结算时拒绝沉默的 child，以此强制该义务。** 没有什么可以拒绝：当结算可被观察时 child 的循环已经结束，让它的拆卸失败只会毁掉工作而不会送达结果。由运行时无条件投递终止事实才是这一情形的答案，而它属于继续执行管理器，不属于本包。
|
||||
|
||||
## 后果
|
||||
|
||||
- 加载本包后，每个可继续进程内 child 的每次请求都会多出一个提示词 section 和一段更长的 `report` 描述；其他任何 Agent 的请求都不变。
|
||||
- 默认部署会为每条被接受的报告唤醒 parent 一次。频繁上报的嵌套树会消耗额外的 parent 轮次；`quiet` 是有文档记载的退路。
|
||||
- `installReportTool` 需要 child 作用域中的 `ctx.systemPrompt`，因此本包在 `inject` 中声明 `systemPrompt`，从而在加载时失败，而不是等到下一次 child 物化时。
|
||||
- 单元覆盖固定了新默认值、两处关键指令措辞、该 section 相对 parent 与同级均仅限 child 的作用域，以及两项注册在安装回滚或撤销时的清理。
|
||||
- 三个带可继续 child 的整体组装 ACP 场景通过新的 sidecar 逐字固定完整的 child 提示词；今后任何对 child 作用域 section 的改动都会让这些场景失败，而不是悄悄通过。
|
||||
|
||||
### 已接受的风险
|
||||
|
||||
默认唤醒会在深层树中放大模型工作量。部署通过 `reportDelivery` 掌握该取舍，且放大幅度以每条被接受报告一个轮次为界。
|
||||
|
||||
child 仍可能不上报就结束，本次改动无法检测这一点。只有运行时自己的[结算记账](2026-08-06-manager-owned-subagent-settlement-delivery.md)才能补上这一情形。
|
||||
@@ -0,0 +1,6 @@
|
||||
# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-06-manager-owned-subagent-settlement-delivery.md
|
||||
2026-08-06-manager-owned-subagent-settlement-delivery.md: e06fe2b4ca7dd9a975f70524979de09d13cdcf79
|
||||
2026-08-06-manager-owned-subagent-settlement-delivery.zh.md: 77e8b5e4bae9b05e08bf2c4a4bf997b688cd1157
|
||||
@@ -0,0 +1,103 @@
|
||||
# Agent Note: Settlement delivery belongs to the continuation manager
|
||||
|
||||
Status: implemented
|
||||
|
||||
English | [中文](2026-08-06-manager-owned-subagent-settlement-delivery.zh.md)
|
||||
|
||||
## Problem
|
||||
|
||||
Continuable background delegation was the one asynchronous operation a model could start but could not reach the end of. Every other shape has a retrieval primitive or a return value: a background bash command and a one-shot background subagent both settle through a Task that `task_output(wait: true)` can block on, a workflow and a foreground subagent return their result to the caller. A continuable background child returned only its durable id, and nothing existed that a parent could wait on or would be handed.
|
||||
|
||||
[The report obligation](2026-08-06-continuable-child-report-obligation.md) closed the cooperative half of that gap by instructing the child to report before it finishes. Instruction cannot close the rest. A child stopped by a token ceiling, a model failure, cancellation, or teardown never reaches the point where it could comply — not rarely, but never — and those are precisely the endings a waiting parent most needs to hear about. The observable downstream symptoms were parents busy-polling `list_agents`, re-sending messages to children that had already settled, and deployments abandoning `subagent` for `workflow` because a workflow at least returns something.
|
||||
|
||||
The signal already existed. `subagent/end` has carried `stopReason` and `lastAssistantMessage` since continuable Activations shipped. What was missing was any consumer that turned it into context the parent's model could see.
|
||||
|
||||
## Decision
|
||||
|
||||
The continuation manager delivers the account itself, from inside the disposal transaction that ends the Activation.
|
||||
|
||||
When a resident Activation settles, `notifySettlement()` resolves the child's durable direct parent and sends it one user-role message: the epoch's outcome as a sentence the parent can act on, then the child's final assistant content, or a statement that it produced none. Delivery is unconditional for every child whose id a caller actually received. It does not consult whether the child reported, and it keeps no bookkeeping that could make the promise conditional — that unconditionality is what lets `tool-subagent` tell the model "you are told when it finishes, so never poll or wait on it" and have that be true. A materialization rolled back before its first accepted message stays silent, because the caller was told that child was not established.
|
||||
|
||||
### Provenance
|
||||
|
||||
The notice carries `{ kind: 'subagent-settled', form: 'notice', summary, senderSessionId }`. It is deliberately not the existing `subagent-report` kind. A report is content the child chose; this is the runtime stating what became of the child. Merging them would credit the child with words it never wrote, and would make a durable log unable to distinguish "the child said it was done" from "the harness observed that it stopped". The `notice` form also gives a UI the collapsed one-line presentation this message wants, where `relay` would present it as correspondence.
|
||||
|
||||
### Two ordering rules, and why the manager owns them
|
||||
|
||||
An external `ctx.on('subagent/end')` listener looks more decoupled and is wrong. `SubagentRunEndInfo` names no parent, the child handle is already disposed when the edge fires so the parent cannot be recovered from it, and the ownership release that wakes the parent's own settlement watcher has already run. The manager holds the parent reference throughout disposal, so none of those obstacles exist for it.
|
||||
|
||||
**The send happens before `releaseOwnership`.** At that point the parent still counts this child, so `stateOf(parent)` is `waiting` and the parent is structurally unable to be judged settled. Delivering after the release instead races a watcher that resumes one microtask later, finds itself childless and quiet, and disposes an Agent whose `cancel()` clears the very inbox the notice is sitting in. The failure mode is a silently missing message with no error anywhere.
|
||||
|
||||
**A resident parent receives it through `admitWaking`.** Registering the message id before the synchronous send is what keeps the window between `followup()` and the microtask that admits it from being read as quiescence. This is not belt-and-braces over the first rule: `Agent.status` folds context maintenance into `idle`, and a waking send behind maintenance only arms a deferred wake, so a parent compacting its context is judged quiet by both `status` and the owned-child set the moment the release lands.
|
||||
|
||||
Both rules are pinned by tests that fail when the ordering is reversed or the accounting removed.
|
||||
|
||||
### Scheduling
|
||||
|
||||
An idle parent gets one ordinary later turn. A busy parent is steered into its nearest step boundary, because `Inbox.claim()` takes the whole next-step batch at one boundary: four children settling together then cost one step rather than four turns. Steering rather than injecting is deliberate — the wake is a no-op while the driver is running, and it closes the window where a driver retires between the status read and the send, which would strand the notice unclaimed until something unrelated woke the parent. This is a correctness rule, not a deployment preference, so it is not a `Config` field.
|
||||
|
||||
One `running` parent is not steerable: one whose turn is already cancelled but has not yet exited. `Agent.send()` redirects waking input submitted after cancellation to the next turn, latches the wake, and replays it once the cancelled driver converges — except for a disposal cancellation, which never latches and belongs to the teardown rule below. The notice therefore still opens its own turn without waiting for unrelated input; the cost is a redirected turn boundary, not the message.
|
||||
|
||||
**A parent whose own teardown began gets no wake.** Waking is not a queue operation: `Agent.followup()` on a quiescent Agent starts a turn, and `cancel()` on an idle Agent is a documented no-op that does not arm against a later one. Every teardown path therefore ends with a live, cancelled, still-registered parent — `drainContinuableDescendants()` is called by the ACP bridge between cancelling its session agents and disposing them — so an unguarded notice starts a real model request on an Agent about to be destroyed, once per tree layer, because each layer's own notice then wakes the layer above it. `notifySettlement()` asks the same question `assertAdmitting()` asks (is this lineage's continuable admission closed?) and injects instead. Injection is not a durable mailbox — Accepted risks records what the parent's own disposal then does to it — but it is the only send that reaches a parent still reading its inbox without arming a turn on one that is not, and nothing is lost that the wake would have delivered: the turn a wake started was itself disposed mid-flight.
|
||||
|
||||
Delivery never blocks or fails teardown. A rejected send is logged and dropped, because retaining a child to retry a notice would pin its whole ancestry in `waiting` forever, and a parent that has left the registry is an ordinary outcome rather than an error.
|
||||
|
||||
### The epoch's own log is the whole account
|
||||
|
||||
`epochStopReason()` reads the epoch's outcome from its own log, because teardown succeeding says nothing about whether the model errored, hit its ceiling, or was stopped. Reading turns alone got that wrong twice, in the same shape both times: a turn stopped before its first step leaves a `turn/end` indistinguishable from the balanced no-op turns a rejection or an emptied claim produces, so the filter that skipped those also skipped real endings and answered with the previous turn's clean completion. The durability checkpoint (`dsh-session-checkpoint-policy`, in every shipped profile) and prompt assembly both run at that boundary and both propagate, and `Inbox.claim()` has already taken the messages by then — so the parent was told a child finished while the delivery it was waiting on had been swallowed. Under a promise that says "you are told when it finishes, so never poll", that is the one failure a parent cannot detect and will not retry.
|
||||
|
||||
The missing fact was never the turn's; it was the inbox's. `Inbox` logs every mutation with `removedCount` and marks a cancellation `outcome: 'canceled'`, which separates a turn claiming its input from work being dropped unrun. `foldConsumedWork()` in `dsh-agent` folds both vocabularies into one answer: the latest turn that accounts for consumed work — stepped, or claimed-then-failed, stopped, or rejected — and whether accepted work was cancelled after it with no turn opening over it. A `blocked` end over claimed input is an account too: the pre-step rejection that produced it — a hook deny, a policy plugin — discarded the messages the turn claimed, so the notice says the child declined rather than finished. Only a `blocked` turn that claimed nothing stays invisible.
|
||||
|
||||
Deriving it from the log rather than from live state is what makes it whole. An earlier version sampled the manager's own Activation immediately before cancelling, which could only ever see cancellations this manager was about to perform: an ancestor's `interrupt()`, or an unloading plugin cancelling an agent it tracks, left the sample false and the notice still saying `finished`. It also left the accepted-but-never-claimed case pinned to nothing a test could distinguish from its absence. One fold over the log covers every issuer, and both halves fail their own tests when removed.
|
||||
|
||||
Precedence is the consumer's: a recorded failure or ceiling wins over a cancellation, because stopping a child that had already failed does not turn its failure into a cancellation. `dsh-agent` owns the fold because it owns the inbox marker the answer depends on, and both consumers already depend on it — the continuable epoch here, and the one-shot `readResult()`, which had the same hole.
|
||||
|
||||
Both matter past the notice: `subagent/end` carries `stopReason` to the jsonrpc UI and the Claude hook bridge, which reported a torn-down mid-turn child as `completed`.
|
||||
|
||||
### Snapshot coverage
|
||||
|
||||
Three assembled ACP scenarios cover the notice: a child that never reports, a child that reports first, and a child driven through several follow-up turns. All three needed an explicit fence. The notice arrives once the child's teardown finishes, which races whatever the parent is already doing, so each scenario holds the child behind the parent's spawn turn and then waits for the parent turn the notice opens (`waitForTurnStart` at that turn, then `waitForTurnEnd`) before the script continues. Waiting for a turn the run is not fenced to produce is not coverage: it is a timeout when the notice lands in the turn already running instead.
|
||||
|
||||
`subagent-continuable` is the one that pins a failure. Its child's last turn dies on the forced durability checkpoint without entering a step, so that transcript is where the stop-reason rule above is visible end to end: the notice says the child *failed*, carries the earlier `SECOND_OK` as its last content rather than as a result, and the parent's own acknowledgement turn reaches the ACP client.
|
||||
|
||||
A keyless headless Loader snapshot covers the user-visible path end to end. Its replay parent starts one continuable child with `run_in_background: true`, never calls `list_agents`, `send_message`, or Task tools, consumes the manager-authored `subagent-settled` notice, and produces its final answer. The child never calls `report`, so the transcript cannot pass through the cooperative report path. A test-only Loader fence holds the parent's post-spawn request until the real manager notice enters its inbox, removing platform scheduling from the transcript without synthesizing the notice.
|
||||
|
||||
`subagent-report` needed one more concession. With the shipped waking report default, that scenario has two independent parent wakes — the report and the settlement — and whether the second extends the first's turn or opens its own is a genuine coin flip that measured 50/50 across runs. No authored transcript can hold both orders. Its overlay therefore pins `reportDelivery: quiet`, leaving settlement as the only wake, and a snapshot-only pre-step fence holds the child until the parent's spawn turn ends so that wake opens one deterministic turn claiming both messages. The waking report default keeps its coverage in the report package's own tests.
|
||||
|
||||
The refusal and interruption wordings are pinned verbatim in unit tests rather than in a replayed transcript: producing them needs a rejecting policy plugin or a cancellation fenced at a step boundary, which the keyless assemblies do not otherwise carry, and the assembled scenarios already pin the notice pathway itself end to end.
|
||||
|
||||
## Alternatives considered
|
||||
|
||||
**Give continuable children a Task.** A Task is a one-shot contract: one producer, one settlement, one result. An Activation runs many turns, outlives any single one, and can be resumed after it ends. Wrapping it in a Task recreates exactly the lifetime mismatch continuable children were introduced to remove, and would make one turn look terminal.
|
||||
|
||||
**Attach an external `subagent/end` listener.** Rejected on three counts above — no parent in the payload, a disposed child handle, and an ordering the listener cannot influence. A listener would also have to be strictly synchronous to beat the release, and nothing at that seam enforces it, so the correct version would be correct only by accident.
|
||||
|
||||
**Deliver only when the child did not report.** This was the first design. It needs per-Activation bookkeeping, still misses the child that reported progress and then died before its result, and — decisively — makes the parent-facing promise conditional. "Usually you are told" is not a contract a tool description can state, and a model that cannot rely on the notice will poll anyway.
|
||||
|
||||
**Make delivery configurable.** A deployment switch would return the model-facing text to "usually", which is the failure this change exists to remove. Protocol constants and safety invariants stay fixed; this is one of them.
|
||||
|
||||
**Change `subagent/end` to carry the parent, and let a plugin deliver.** That widens a published payload for one in-package consumer, keeps every ordering hazard, and makes the return channel an optional plugin again. Extending the package-private `ActivationObserver` with `terminal(failure)` keeps one computation of the terminal facts and no public surface change.
|
||||
|
||||
**Always use `followup`.** Simpler and uniform, but a fan-out of children settling together would cost one parent turn each. The step-boundary batch already exists; using it is free.
|
||||
|
||||
## Consequences
|
||||
|
||||
- A continuable child's parent receives one message per settled Activation. Fan-out deployments therefore add parent turns; steering keeps a simultaneous batch to one step.
|
||||
- `tool-subagent` promises the notice in its schema because the return channel is service behavior, not an optional plugin.
|
||||
- `Activation` carries `parentSession` and `announced`. The first exists because the child handle is disposed before delivery; the second is what keeps a rolled-back materialization silent.
|
||||
- `foldConsumedWork()` replaces `dsh-session`'s `findLastMessageTurnEnd()` and moves to `dsh-agent`, which owns the inbox marker it reads; the one-shot in-process path folds the same answer and does not classify a cut-short one-shot child as `completed`.
|
||||
- Unit coverage pins the unconditional contract, each terminal reason, idle and busy scheduling, the batch, the maintenance regression, the pre-release ordering, a parent that is gone, and a rejected send that must not fail teardown.
|
||||
- Three ACP scenarios use an explicit settlement fence, and `subagent-report` has a config overlay that pins quiet report delivery.
|
||||
- A keyless headless Loader snapshot pins background start → manager-authored settlement notice → final parent answer with no polling or child `report` call.
|
||||
|
||||
### Accepted risks
|
||||
|
||||
The notice is delivered, not confirmed. There is no durable mailbox, receipt, or retry: a parent that is not live loses it, and the child's Session remains the only durable record. Closing that needs an offline mailbox protocol with its own addressing, authorization, and replay rules.
|
||||
|
||||
A notice injected during teardown is not read by a model when that parent is disposed next, which every teardown caller does: the disposal cancel clears the unclaimed message and the log keeps the insert/cancel pair as the record. Making teardown delivery readable after resume requires either the offline mailbox above or a change to disposal of durable pending work. Disposal discards every unclaimed inbox item, including user input, so changing that behavior is a core-agent decision rather than a settlement-delivery detail. After resume, the parent can discover the child but does not receive the outcome: `list_agents` reports existence and live-or-stored status only — `SubagentListEntry.activity` says so — and recovering the ending requires asking the child through `send_message`.
|
||||
|
||||
Stop-reason attribution is a best effort over the log's existing splice vocabulary, biased against overstating success. `Inbox.remove()` and teardown's `clear()` write identical cancellation splices, so removing a message whose content survives elsewhere — `workspace-context` vacuuming a pending instruction refresh, or settlement's own cancel clearing one left pending — can read as work dropped unrun and report a finished child as stopped. Separating them requires a richer removal vocabulary in `dsh-agent`; without it, the misread is narrow and errs toward the parent double-checking a finished child, never toward trusting an unfinished one.
|
||||
|
||||
Turn amplification is real for deep or wide trees, and it is not configurable by design. The step-boundary batch bounds it for simultaneous settlement but not for children that settle apart.
|
||||
|
||||
Two independent waking sources cannot be ordered in an authored transcript. The assembled coverage pins each separately rather than their interleaving.
|
||||
@@ -0,0 +1,103 @@
|
||||
# Agent Note: Settlement delivery belongs to the continuation manager
|
||||
|
||||
Status: implemented
|
||||
|
||||
[English](2026-08-06-manager-owned-subagent-settlement-delivery.md) | 中文
|
||||
|
||||
## Problem
|
||||
|
||||
可继续后台委派是模型唯一一种能够发起、却无法抵达终点的异步操作。其他每一种形态都有取回原语或返回值：后台 bash 命令与一次性后台 subagent 都通过 Task 结算，`task_output(wait: true)` 可以阻塞等待；workflow 与前台 subagent 会把结果返回给调用方。可继续后台 child 只返回它持久化的 id，而父级既没有可等待的对象，也不会被交付任何东西。
|
||||
|
||||
[报告义务](2026-08-06-continuable-child-report-obligation.md)通过要求 child 在结束前上报，补上了这一缺口中协作的那一半。指令无法补上其余部分。被 token 上限、模型失败、取消或拆卸终止的 child 永远走不到能够遵守的那一步——不是很少，而是从不——而这些恰恰是等待中的父级最需要被告知的结束方式。可观察到的下游症状包括：父级忙轮询 `list_agents`、向已经结算的 child 反复发送消息，以及部署放弃 `subagent` 转用 `workflow`，因为 workflow 至少会返回点什么。
|
||||
|
||||
信号本身早就存在。自可继续 Activation 发布以来，`subagent/end` 就一直携带 `stopReason` 与 `lastAssistantMessage`。缺的是把它变成父级模型能看到的上下文的那个消费者。
|
||||
|
||||
## Decision
|
||||
|
||||
继续执行管理器自己投递这份记账，就在结束 Activation 的那笔 dispose 事务内部完成。
|
||||
|
||||
当驻留 Activation 结算时，`notifySettlement()` 解析该 child 持久化的直接父级，并向它发送一条用户角色消息：先是父级可据以行动的一句结果说明，然后是 child 的最终 assistant 内容，或一句说明它没有产出内容。对每个调用方真正拿到过 id 的 child，投递都是无条件的。它不查询 child 是否上报过，也不保留任何可能让这项承诺变成有条件的记账——正是这种无条件性，才让 `tool-subagent` 能够告诉模型「它结束时你会被告知，因此绝不要轮询或等待它」并且这句话为真。在第一条消息被接受之前就回滚的物化保持静默，因为调用方已被告知该 child 未建立。
|
||||
|
||||
### 来源信息
|
||||
|
||||
该通知携带 `{ kind: 'subagent-settled', form: 'notice', summary, senderSessionId }`，刻意不复用既有的 `subagent-report` kind。上报是 child 选择的内容；这条消息则是运行时在陈述这个 child 后来怎样了。把两者合并会把 child 从未写过的话算到它头上，也会让持久化日志无法区分「child 说它做完了」和「harness 观察到它停下了」。`notice` 形态还为 UI 提供了这条消息想要的折叠单行呈现，而 `relay` 会把它呈现为往来信件。
|
||||
|
||||
### 两条顺序规则，以及为什么归管理器所有
|
||||
|
||||
外部 `ctx.on('subagent/end')` listener 看起来更解耦，但它是错的。`SubagentRunEndInfo` 不指名父级；该边触发时 child handle 已被 dispose，因此无法从中恢复父级；而唤醒父级自身结算 watcher 的所有权释放也已经执行过了。管理器在整个 dispose 过程中都持有父级引用，因此这些障碍对它都不存在。
|
||||
|
||||
**发送发生在 `releaseOwnership` 之前。** 此刻父级仍然计入这个 child，因此 `stateOf(parent)` 为 `waiting`，父级在结构上不可能被判定为已结算。改在释放之后投递，则会与一个在下一个 microtask 恢复的 watcher 竞争：它会发现自己没有 child 且处于静止，于是 dispose 一个 Agent，而该 Agent 的 `cancel()` 会清空正装着这条通知的那个 inbox。失效表现是一条静默丢失的消息，任何地方都不会报错。
|
||||
|
||||
**驻留父级通过 `admitWaking` 接收它。** 在同步发送之前登记消息 id，正是让 `followup()` 与承认它的那个 microtask 之间的窗口不被读作静止的原因。这不是对第一条规则的多余保险：`Agent.status` 会把上下文维护折叠成 `idle`，而维护期间的唤醒发送只会预置一次延后唤醒，因此正在压缩上下文的父级，在所有权释放落地的那一刻会同时被 `status` 与已拥有 child 集合判定为静止。
|
||||
|
||||
两条规则都有测试固定：把顺序反转或去掉记账，测试就会失败。
|
||||
|
||||
### 调度
|
||||
|
||||
空闲父级得到一个普通的后续轮次。繁忙父级则被 steer 到其最近的 step 边界，因为 `Inbox.claim()` 会在一个边界上整批取走 next-step：四个 child 同时结算时因此只消耗一个 step，而不是四个轮次。采用 steer 而非 inject 是刻意的——驱动运行期间该唤醒是空操作，同时它关闭了「驱动在状态读取与发送之间退出」的那个窗口；否则通知会滞留无人认领，直到别的事件唤醒父级。这是正确性规则而非部署偏好，因此不做成 `Config` 字段。
|
||||
|
||||
有一种 `running` 父级是无法 steer 的：轮次已被 cancel 但尚未退出的那种。`Agent.send()` 会把取消之后提交的唤醒输入改投到下一个轮次、闩存这次唤醒，并在被取消的驱动收敛后重放它——只有 disposal 取消从不闩存，那属于下面的拆卸规则。因此通知仍会开启自己的轮次，无需等待无关输入；代价是一次被改投的轮次边界，而不是消息本身。
|
||||
|
||||
**自身已开始拆卸的父级不会被唤醒。** 唤醒不是入队操作：对静息 Agent 调用 `Agent.followup()` 会开启一个轮次，而对空闲 Agent 调用 `cancel()` 是文档明确的空操作，不会对之后的轮次设防。因此每条拆卸路径最终都面对一个在线、已取消、仍在注册表中的父级——ACP 桥接层正是在取消其 session agent 与 dispose 它们之间调用 `drainContinuableDescendants()`——于是一条无防护的通知会在一个即将被销毁的 Agent 上发起真实模型请求，而且每层树各一次，因为每层自己的通知又会唤醒它上面那层。`notifySettlement()` 会问 `assertAdmitting()` 问的同一个问题（这条谱系的可继续准入是否已关闭？），并改为 inject。inject 不是持久 mailbox——父级自身的 dispose 会对它做什么，记在「已接受的风险」里——但它是唯一能送达仍在读取自身 inbox 的父级、又不会在不该被唤醒的父级上预置一个轮次的发送方式；而唤醒本可送达的东西一样没有丢失：唤醒开启的那个轮次本身就会在半途被 dispose。
|
||||
|
||||
投递绝不会阻塞或使拆卸失败。发送被拒会被记录并丢弃，因为为重试一条通知而保留 child，会把它的整条祖先链永久钉在 `waiting` 上；而父级已离开注册表属于普通结果，不是错误。
|
||||
|
||||
### epoch 自己的日志就是全部交代
|
||||
|
||||
`epochStopReason()` 从 epoch 自己的日志读取结局，因为拆卸成功与否，对「模型是否报错、是否撞到上限、是否被停下」什么也没说明。只读轮次这件事已经错了两次，而两次的形状相同：在第一个 step 之前被停下的轮次，其 `turn/end` 与「拒绝」或「被清空的认领」产生的平衡空转轮次长得一模一样，于是那道用来跳过后者的过滤，也把真实的结局一起跳过了，转而用上一个轮次的干净收尾作答。持久化检查点（`dsh-session-checkpoint-policy`，存在于每个随附 profile 中）与提示词组装都运行在这个边界上、且都会向外传播，而此时 `Inbox.claim()` 已经把消息取走了——于是父级被告知 child 已完成，而它正在等待的那条投递已被吞掉。在「你会在它完成时被告知，所以永远不要轮询」这一承诺之下，这恰恰是父级无法察觉、也不会重试的那一种失败。
|
||||
|
||||
缺失的事实从来不属于轮次，而属于 inbox。`Inbox` 会把每次改动连同 `removedCount` 一起记入日志，并给取消标记 `outcome: 'canceled'`，这就把「某个轮次认领了它的输入」与「工作被丢弃且从未运行」区分开来。`dsh-agent` 中的 `foldConsumedWork()` 把两套词汇折叠成一个答案：能为已消费工作作出交代的最新轮次——进入过 step 的，或认领后失败、被停下或被拒绝的——以及此后是否有已接受的工作被取消、而没有任何轮次为它开启过。认领过输入、以 `blocked` 结束的轮次同样是一份交代：产生它的 pre-step 拒绝——hook deny、策略插件——把该轮次认领的消息一并丢弃了，因此通知会说 child 拒绝了任务，而不是完成了任务。只有没认领任何输入的 `blocked` 轮次保持不可见。
|
||||
|
||||
从日志而不是从活动状态推导，才让它完整。早先的版本会在 cancel 之前立刻采样管理器自己的 Activation，而那样只能看到本管理器即将执行的取消：来自祖先的 `interrupt()`，或某个正在卸载的插件取消它所跟踪的 Agent，都会让该采样为假，通知照旧说 `finished`。它也让「已接受但从未被认领」这一情形没有任何测试能把它与「该判据不存在」区分开。一次对日志的折叠覆盖了所有发起方，而两个半边在被移除时都会让各自的测试失败。
|
||||
|
||||
优先级归消费方：已记录的失败或上限优先于取消，因为停下一个已经失败的 child，不会把它的失败变成一次取消。`dsh-agent` 拥有这个 fold，是因为答案所依赖的那个 inbox 标记归它所有，而两个消费方本来就依赖它——这里的可继续 epoch，以及一次性的 `readResult()`（它有同一个漏洞）。
|
||||
|
||||
两者的影响都超出通知本身：`subagent/end` 会把 `stopReason` 送到 jsonrpc UI 与 Claude hook 桥接层，而它们此前把被拆卸的、正在跑轮次的 child 报成 `completed`。
|
||||
|
||||
### 快照覆盖
|
||||
|
||||
三个整体组装的 ACP 场景覆盖该通知：一个从不上报的 child、一个先上报的 child，以及一个被多轮 follow-up 驱动的 child。三者都需要显式栅栏。通知在 child 拆卸完成后才到达，会与父级当时正在做的事竞争，因此每个场景都会把 child 保持到父级启动轮次结束，再等待该通知开启的那个父级轮次（先 `waitForTurnStart` 到该轮次，再 `waitForTurnEnd`），然后脚本才继续。等待一个运行并未被栅栏保证会产生的轮次不算覆盖：一旦通知落进已经在跑的那个轮次，它就是一次超时。
|
||||
|
||||
`subagent-continuable` 是其中固定失败结局的那个。它的 child 最后一个轮次在被强制的持久化检查点上死亡，且未进入任何 step，因此该 transcript 正是上面那条终止原因规则的端到端可见之处：通知说该 child **失败**，把此前的 `SECOND_OK` 作为它最后产出的内容而非结果携带，而父级自己的确认轮次会到达 ACP 客户端。
|
||||
|
||||
另有一个无密钥的 headless Loader 快照端到端覆盖用户可见路径。其重放父级通过 `run_in_background: true` 启动一个可继续 child，从不调用 `list_agents`、`send_message` 或 Task 工具，消费管理器写入的 `subagent-settled` 通知，并给出最终答案。child 从不调用 `report`，因此该 transcript 不可能经由协作式上报路径通过。一个仅用于测试的 Loader 栅栏会把父级启动后的请求保持到真实管理器通知进入其 inbox 为止，从 transcript 中排除平台调度差异，但不会伪造该通知。
|
||||
|
||||
`subagent-report` 还需要多做一步让步。在随附的唤醒上报默认值下，该场景有两个互相独立的父级唤醒——上报与结算——而第二个究竟是延长第一个的轮次还是另开一个轮次，是一枚真正的硬币，多次运行实测约为五五开。任何手写 transcript 都无法同时容纳两种顺序。因此它的 overlay 固定 `reportDelivery: quiet`，使结算成为唯一唤醒；另一个仅用于快照的 pre-step 栅栏会把 child 保持到父级启动轮次结束，使这次唤醒开启一个确定轮次并同时认领两条消息。唤醒上报默认值的覆盖则保留在 report 包自身的测试中。
|
||||
|
||||
拒绝与中断两种措辞在单元测试中逐字钉死，而不进入重放 transcript：触发它们需要一个会拒绝的策略插件、或一次在 step 边界被栅栏卡住的取消，而无密钥组装本身并不携带这些；通知通路本身已由整体组装场景端到端钉住。
|
||||
|
||||
## Alternatives considered
|
||||
|
||||
**给可继续 child 引入 Task。** Task 是一次性契约：一个生产者、一次结算、一个结果。Activation 会执行许多轮次、比其中任何一轮活得更久，并且可以在结束后被恢复。用 Task 包装它，恰好重建了可继续 child 当初为消除而引入的生命周期错配，还会让某一个轮次看起来是终局。
|
||||
|
||||
**挂一个外部 `subagent/end` listener。** 因上文三点被否决——payload 里没有父级、child handle 已被 dispose，以及 listener 无法影响的顺序。listener 还必须严格同步才能抢在释放之前，而该 seam 上没有任何东西强制这一点，因此正确的版本只能靠碰巧正确。
|
||||
|
||||
**仅在 child 没有上报时投递。** 这是最初的设计。它需要按 Activation 记账，仍会漏掉「报了进度、随后在给出结果前死掉」的 child，而且最关键的是：它让面向父级的承诺变成有条件的。「通常你会被告知」不是工具描述能陈述的契约，而无法依赖该通知的模型无论如何都会去轮询。
|
||||
|
||||
**把投递做成可配置。** 部署开关会把面向模型的文本重新变回「通常」，而这正是本次改动要消除的失效。协议常量与安全不变量保持固定；这就是其中之一。
|
||||
|
||||
**修改 `subagent/end` 让它携带父级，由插件负责投递。** 那会为一个包内消费者拓宽已发布的 payload，保留全部顺序风险，并让返回通道重新变成可选插件。以 `terminal(failure)` 扩展包私有的 `ActivationObserver`，则只保留一处终止事实的计算，且不改动任何公开面。
|
||||
|
||||
**始终使用 `followup`。** 更简单也更统一，但一批同时结算的 child 会各自消耗一个父级轮次。step 边界的批量语义本来就存在，用它是免费的。
|
||||
|
||||
## Consequences
|
||||
|
||||
- 可继续 child 的父级会为每个已结算 Activation 收到一条消息。因此，做扇出的部署会增加父级轮次；steer 会把同时结算的一批压缩到一个 step。
|
||||
- `tool-subagent` 在其 schema 中承诺该通知，因为返回通道是服务行为，不是可选插件。
|
||||
- `Activation` 携带 `parentSession` 与 `announced`。前者存在是因为 child handle 在投递前已被 dispose；后者让被回滚的物化保持静默。
|
||||
- `foldConsumedWork()` 取代 `dsh-session` 的 `findLastMessageTurnEnd()`，并迁移到 `dsh-agent`——它拥有该 fold 所读取的 inbox 标记；一次性 in-process 路径折叠同一个答案，不会把被中途切断的一次性 child 归类为 `completed`。
|
||||
- 单元覆盖固定了无条件约定、每种终止原因、空闲与繁忙两种调度、批量语义、维护期回归、释放前顺序、父级已消失，以及一次不得让拆卸失败的发送被拒。
|
||||
- 三个 ACP 场景使用显式的结算栅栏，`subagent-report` 带有固定静默上报投递的配置 overlay。
|
||||
- 一个无密钥的 headless Loader 快照固定了「后台启动 → 管理器写入的结算通知 → 父级最终答案」路径，其中没有轮询，也没有 child `report` 调用。
|
||||
|
||||
### 已接受的风险
|
||||
|
||||
通知只是被投递，而不是被确认。没有持久化 mailbox、回执或重试：不在线的父级会丢失它，child 的 Session 仍是唯一的持久记录。要补上这一点，需要一套带有自身寻址、授权与重放规则的离线 mailbox 协议。
|
||||
|
||||
当父级紧接着被 dispose 时（每个拆卸调用方都会这么做），在拆卸期间被 inject 的通知不会被模型读到：dispose 的 cancel 会清除这条未被认领的消息，而日志保留 insert/cancel 这一对作为记录。要让拆卸期投递在 resume 之后仍可读，要么需要上面那套离线 mailbox，要么需要改变 dispose 对持久待处理工作的处理方式。dispose 会丢弃每一条未被认领的 inbox 项，用户输入也不例外，因此改变该行为是一个 core-agent 决策，而不是结算投递的细节。resume 后的父级可以发现 child，但不会收到结局：`list_agents` 只报告存在性与「在线/仅存储」状态——`SubagentListEntry.activity` 就是这么写的——要取回结局，必须通过 `send_message` 去问那个 child。
|
||||
|
||||
终止原因的归因是对日志既有 splice 词汇的尽力而为，偏向永不高估成功。`Inbox.remove()` 与拆卸的 `clear()` 写出的取消 splice 完全相同，因此删除一条内容仍保留在别处的消息——`workspace-context` 清理待处理的 instructions 刷新、或结算自身的 cancel 清掉一条仍在挂起的这类消息——可能被读作「工作被丢弃且从未运行」，把已完成的 child 报成被停下。区分二者需要 `dsh-agent` 提供更丰富的删除词汇；在该词汇可用前，这项误读的范围很窄，且错的方向是让父级复查一个已完成的 child，而永远不是信任一个未完成的 child。
|
||||
|
||||
对于深或宽的树，轮次放大是真实存在的，而且按设计不可配置。step 边界的批量语义只能约束同时结算的情形，无法约束分散结算的 child。
|
||||
|
||||
两个互相独立的唤醒源无法在手写 transcript 中排序。整体组装覆盖分别固定它们，而不固定它们的交错。
|
||||
@@ -2,5 +2,5 @@
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-08-windows-acl-restricted-token-sandbox.md
|
||||
2026-08-08-windows-acl-restricted-token-sandbox.md: 7e8f229269233d9ac9baa65241ca02a4cf4c3f7c
|
||||
2026-08-08-windows-acl-restricted-token-sandbox.zh.md: eeb346b228b3559f487448e5d4ec525b7bb89525
|
||||
2026-08-08-windows-acl-restricted-token-sandbox.md: 972713e02860218853f421aa700a8b60b33ada5b
|
||||
2026-08-08-windows-acl-restricted-token-sandbox.zh.md: da41cb3f9aa46bab96a5fbb6c035205b22c442aa
|
||||
|
||||
@@ -6,15 +6,15 @@ English | [中文](2026-08-08-windows-acl-restricted-token-sandbox.zh.md)
|
||||
|
||||
## Problem
|
||||
|
||||
The [sandbox decision](2026-07-06-sandbox.md) leaves `PLATFORM_CHAINS.win32` empty, so shipped Windows profiles degrade to danger-full-access because no confining executor exists. The win32 rung must confine the two file-effect modes the sandbox vocabulary promises — `read-only` (zero writes) and `workspace-write` (writes under the workspace root plus a backend-defined temp area) — while leaving reads, network, and process visibility alone, because every mode permits reading.
|
||||
The original [sandbox decision](2026-07-06-sandbox.md) left `PLATFORM_CHAINS.win32` empty, so shipped Windows profiles degraded to danger-full-access because no confining executor existed. The win32 rung must govern the two file-effect modes in the sandbox vocabulary — `read-only` (no explicit writable root) and `workspace-write` (writes under the workspace root plus a backend-defined temp area) — while reporting any effects its mechanism cannot govern; reads, network, and process visibility remain outside this vocabulary.
|
||||
|
||||
## Decision
|
||||
|
||||
Implement the rung directly on the raw ACL mechanism: duplicate the caller's token into a `WRITE_RESTRICTED` token (`CreateRestrictedToken` with `WRITE_RESTRICTED` + `DISABLE_MAX_PRIVILEGE` + `LUA_TOKEN`) whose restricting SIDs include a write SID (`S-1-4-x-y`); the write SID's Write ACEs on the workspace and temp roots are the entire write allowlist, because `WRITE_RESTRICTED` intersects write accesses only and reads keep the caller's full ambient access. The mechanism is the one huoyaoyuan/windows-acl-restrict-poc (`10e4dfb`) demonstrates; this port checks every API call and fails closed (the POC fail-opened on every ignored return value). The write SID is the PER-WORKSPACE identity, derived deterministically from the canonical workspace path (`workspaceWriteSid` — sha256 → `S-1-4-x-y`) and stored NOWHERE: the workspace-root ACE therefore materializes once per workspace per machine — the standing ACE is the cross-session reuse cache, and every later provision hits the exact-ACE skip (idempotent re-grant skips the eager full-tree re-propagation — no garbage collection) — instead of once per session, which is what the earlier per-session random SID paid a full tree propagation per session for. The seam derives the session's PRIVATE temp subdirectory from the session id + workspace (sha256, 16 hex — stored nowhere, so no tamper surface exists) and creates it exclusively; it is removed on provider dispose, and a crash leaves it as `%TEMP%` litter whose next resume fails loudly at the exclusive creation until temp hygiene reclaims it. The seam materializes the workspace ACE STANDING (never revoked — the cache) and the temp ACE REVOCABLY (revoked on provider dispose, so an inheritable ACE never outlives its session's temp directory on the ambient temp root). The token's restricting list is the keep-alive group plus the write SID only under workspace-write: read-only = [logon SID, Everyone] and workspace-write = [logon SID, Everyone, write SID]. The keep-alive invariants are logon SID + Everyone (early DLL init dies with 0xC0000142 and CNG crashes pwsh with 0xE0434352 without them). Read-only carries no write SID: a standing grant ACE from an earlier workspace-write period stays INERT (the pass-2 check grants only what the list carries, so read-only remains strictly zero-grant across a `/permission` downgrade or a crash-resumed session, while the standing ACE keeps the re-upgrade free). Authenticated Users is absent from BOTH lists — the WMI namespace security check fails (0x80041003), so CIM is unavailable in every confined mode, and the C:\-root tree-creation escape (standing `AU:(AD)` + `AU:(OI)(CI)(IO)(M)` ACEs) is closed in both; INTERACTIVE/LOCAL are likewise absent from both (the Public tree writes are denied — pinned by the runner's Public-probe regression). Workspace-write children see a PRIVATE per-session temp subdirectory (`<temp>\dsh-<16 hex>` derived from the session id + workspace — created exclusively, reparse points rejected, removed on provider dispose — TMP/TEMP rewritten by the runner — bwrap `--tmpfs /tmp` semantics). The restricted token's DEFAULT DACL is extended with a full-access write-SID ACE (`SetTokenInformation(TokenDefaultDacl)`): new objects created without an explicit security descriptor (anonymous pipes — CreatePipe, sync objects) then carry a restricting-SID ACE and pass the write pass-2 check at creation; NAMED pipes are exempt — their default security descriptor is the Win32 layer's user-mode default SD template (built by KernelBase — owner/SYSTEM/Admins full, Everyone/ANONYMOUS read-only), which the token cannot influence, so piped stdio capture stays denied for confined grandchildren (the POC-documented boundary, pinned by the runner suite). It ships as [`@deepseek-ai/dsh-sandbox-windows-acl`](../../../../packages/sandbox/sandbox-windows-acl/README.md) (backend plus the `./runner` argv-prefix entry), the `win32` chain rung of [`dsh-sandbox-local`](../../../../packages/sandbox/sandbox-local/README.md), and [`@deepseek-ai/dsh-pwsh-sandbox`](../../../../packages/bash/pwsh-sandbox/README.md) as the confining executor; the Windows platform layer re-enables the full permission surface (sandbox/sandbox-policy/permission/approval/fs-sandbox) over the confined pwsh stack.
|
||||
Implement the rung directly on the raw ACL mechanism: duplicate the caller's token into a `WRITE_RESTRICTED` token (`CreateRestrictedToken` with `WRITE_RESTRICTED` + `DISABLE_MAX_PRIVILEGE` + `LUA_TOKEN`) whose restricting SIDs carry distinct workspace and private-temp capabilities. `WRITE_RESTRICTED` intersects write accesses only, so reads keep the caller's ambient access while a write must also match one of these capability ACEs. The mechanism is the one huoyaoyuan/windows-acl-restrict-poc (`10e4dfb`) demonstrates; this port checks every API call and fails closed (the POC fail-opened on ignored failures). The per-workspace SID is derived deterministically from the canonical workspace path (`workspaceWriteSid` — sha256 → `S-1-4-x-y`); its standing workspace ACE is the cross-session reuse cache, and an exact-ACE skip prevents repeated eager tree propagation. Each live session/workspace pair instead gets a random private temp directory and a domain-separated SID derived from that path (`tempWriteSid`); its ACE is revocable, TMP/TEMP point at that directory, and the token default DACL names the temp SID so newly created temp objects do not acquire the shared workspace capability. A fork therefore cannot write its sibling's temp tree. A fresh provider chooses a new path and SID even for the same resumed session, so crash residue is inert litter rather than a collision or inherited capability; agentless calls create and remove the same shape per invocation. The ambient temp root is never an implicit grant. A workspace equal to or containing the temp root fails before any ACL mutation because its inheritable standing ACE would otherwise reach every private child; the direct API rejects overlap in either direction between a writable root and the actual private temp directory. PowerShell can complete its startup AppLocker probe through this private-temp capability, so `workspace-write` remains FullLanguage absent a host-wide policy; `read-only` cannot create the probe files and conservatively enters ConstrainedLanguage. That split is PowerShell startup behavior, not part of the ACL boundary. The token lists are read-only = [logon SID, Everyone] and workspace-write = [logon SID, Everyone, workspace SID, optional temp SID]. Logon SID + Everyone are keep-alive invariants (early DLL init dies with 0xC0000142 and CNG crashes pwsh with 0xE0434352 without them). Because Everyone remains, an external object granting Everyone write access clears both checks; because NTFS ACLs belong to file objects, a granted workspace hard link also grants an external alias. Rejecting all hard links would reject ordinary pnpm workspaces, so the provider reports `enforcement: 'partial'` and the native suite pins both gaps. Read-only carries no capability SID, so standing workspace ACEs remain inert across a mode downgrade. Authenticated Users is absent from both lists — CIM is unavailable, closing the C:\-root tree-creation escape — and INTERACTIVE/LOCAL are absent, denying Public-tree writes. New anonymous pipes and sync objects inherit the temp SID (or workspace SID when temp is disabled, Everyone under read-only) through `SetTokenInformation(TokenDefaultDacl)`; named pipes keep the Win32 layer's owner/SYSTEM/Admins-full, Everyone/ANONYMOUS-read-only template, so piped grandchild stdio remains denied. It ships as [`@deepseek-ai/dsh-sandbox-windows-acl`](../../../../packages/sandbox/sandbox-windows-acl/README.md), the `win32` rung of [`dsh-sandbox-local`](../../../../packages/sandbox/sandbox-local/README.md), and [`@deepseek-ai/dsh-pwsh-sandbox`](../../../../packages/bash/pwsh-sandbox/README.md) as the confining executor.
|
||||
|
||||
## How the restriction works (why no new identity)
|
||||
|
||||
The identity routes restrict by *who* runs the child; this rung restricts by *token derivation*. An identity route (landstrip's restricted-user, AppContainer) runs the child under a fresh account or container SID that starts with zero ACEs on the host's files — everything, reads included, defaults to denied, and every path the child may touch must then be opened back up by writing ACEs for that identity: the wholesale DACL mutation that disqualified both alternatives. The restricted token keeps the caller's own SID and logon session: [`CreateRestrictedToken`](https://learn.microsoft.com/en-us/windows/win32/api/securitybaseapi/nf-securitybaseapi-createrestrictedtoken) derives a token that adds the restricting SIDs and the `WRITE_RESTRICTED` flag, so Windows performs the access check twice — once against the normal SIDs, once against the restricting SIDs — and grants write-class access only where both checks pass. Reads pass on the normal check alone (the caller's SIDs already carry read access everywhere the caller can read), which is why this rung needs no read grants and no new account; writes must additionally clear the orphan-SID check, which only the workspace and temp ACEs satisfy. `DISABLE_MAX_PRIVILEGE | LUA_TOKEN` synthesize the limited-user effect of a fresh account token-side, so even an elevated caller derives a filtered token. The same primitive could restrict reads (`SidsToDisable` turning SIDs deny-only), but a read-restricted token would need per-path read grants — reintroducing exactly the cost the identity routes pay — and the sandbox vocabulary never requires read confinement.
|
||||
The identity routes restrict by *who* runs the child; this rung restricts by *token derivation*. An identity route (landstrip's restricted-user, AppContainer) runs the child under a fresh account or container SID that starts with zero ACEs on the host's files — everything, reads included, defaults to denied, and every path the child may touch must then be opened back up by writing ACEs for that identity: the wholesale DACL mutation that disqualified both alternatives. The restricted token keeps the caller's own SID and logon session: [`CreateRestrictedToken`](https://learn.microsoft.com/en-us/windows/win32/api/securitybaseapi/nf-securitybaseapi-createrestrictedtoken) derives a token that adds the restricting SIDs and the `WRITE_RESTRICTED` flag, so Windows performs the access check twice — once against the normal SIDs, once against the restricting SIDs — and grants write-class access only where both checks pass. Reads pass on the normal check alone (the caller's SIDs already carry read access everywhere the caller can read), which is why this rung needs no read grants and no new account; writes must additionally clear the capability-SID check, which only the workspace and temp ACEs satisfy. `DISABLE_MAX_PRIVILEGE | LUA_TOKEN` synthesize the limited-user effect of a fresh account token-side, so even an elevated caller derives a filtered token. The same primitive could restrict reads (`SidsToDisable` turning SIDs deny-only), but a read-restricted token would need per-path read grants — reintroducing exactly the cost the identity routes pay — and the sandbox vocabulary never requires read confinement.
|
||||
|
||||
## Alternatives considered
|
||||
|
||||
@@ -32,11 +32,11 @@ The [landstrip evaluation](../../rejected/feature/2026-07-26-evaluate-landstrip-
|
||||
|
||||
## Consequences
|
||||
|
||||
Bought: write-only confinement with no new OS floor (`CreateRestrictedToken` predates the mxc releases by two decades), reads/network/process visibility untouched exactly as the mode vocabulary requires, and fail-closed errors carrying the API name and the exact Win32 code. Cost: no read-side or network isolation; console isolation unavailable (hidden-console children die with `STATUS_DLL_INIT_FAILED`; children share the host console); standing ACE mutations on the granted roots (caller-owned directories; workspace ACEs stand forever by design — the reuse cache, invisible residue when a workspace is renamed — temp ACEs revoked by provider dispose together with the derived private temp directory — a crash leaves both behind and the next resume fails loudly at the exclusive creation until temp hygiene reclaims the directory); grant materialization is an EAGER full-tree propagation (`SetNamedSecurityInfoW` walks every descendant immediately — tens of seconds on large workspaces), paid once per workspace per machine by the per-workspace identity; CIM is unavailable in BOTH confined modes (AuthUsers dropped from both lists — the WMI namespace security check fails, and `Get-ComputerInfo` silently returns incomplete results) as the price of closing the C:\-root tree-creation escape in both; FAT-class (non-ACL) targets outside the granted roots remain writable under both modes (no security descriptors to intersect — a legacy residue treated as unsupported, warn-only, documented in the README); NULL-DACL directories are not identity-preserving under a grant+revoke round-trip (documented edge, the POC shares it); `whoami` and token-inspection cmdlets fail under the restricted token (diagnostic noise, documented); and BOTH confined modes run `pwsh` in ConstrainedLanguage mode — the restricted token trips PowerShell's lockdown detection, so `Add-Type`, non-core .NET statics (`[System.IO.*]::`, `[math]::`), COM objects, and reflection fail with "only core types" errors while `-f` formatting, property access, and core cmdlets/types keep working, and the language mode cannot be lifted back to FullLanguage from inside — taught to the model in the pwsh tool description and documented in the package README's Known Limitations; BOTH confined modes also deny named-pipe opens — libuv's piped-stdio spawns fail with EPERM (the POC-documented "no output redirection" boundary; inherited/ignored stdio and anonymous pipes work) — documented in the package README's Known Limitations and taught to the model in the pwsh tool description.
|
||||
Bought: write-only confinement with no new OS floor (`CreateRestrictedToken` predates the mxc releases by two decades), reads/network/process visibility untouched exactly as the mode vocabulary requires, and fail-closed errors carrying the API name and exact Win32 code. Sessions share the intentionally standing workspace capability but not their revocable temp capabilities; restart residue cannot block or authorize a resumed session. Cost: enforcement is structurally partial because Everyone-granted writes and NTFS hard-link aliases cannot be path-confined by this token shape; no read-side or network isolation; console isolation unavailable (hidden-console children die with `STATUS_DLL_INIT_FAILED`; children share the host console); standing workspace ACE mutations (the reuse cache, plus inert residue when a workspace is renamed) and random temp litter after an unclean shutdown until OS hygiene reclaims it; EAGER full-tree workspace propagation (`SetNamedSecurityInfoW` walks every descendant immediately — tens of seconds on large workspaces), paid once per workspace per machine; CIM unavailable in both confined modes (Authenticated Users is absent, closing the C:\-root tree-creation escape); FAT-class non-ACL targets still writable; NULL-DACL directories not identity-preserving under a grant/revoke round trip; `whoami` and token-inspection cmdlets failing under the restricted token; read-only pwsh entering ConstrainedLanguage while workspace-write remains FullLanguage absent host policy; and named-pipe opens remaining denied, so libuv piped-stdio grandchildren fail with EPERM while inherited/ignored stdio and anonymous pipes work. The package README owns these operational limits.
|
||||
|
||||
## Testing
|
||||
|
||||
The product-visible Windows roster flip is win32-only, so the keyless snapshot fixtures — which must replay on macOS/Linux — cannot cover it; the bundle composition specs ([`base.spec.ts`](../../../../packages/bundle/base/tests/base.spec.ts), [`windows-shell.spec.ts`](../../../../apps/cli/tests/windows-shell.spec.ts)) plus the win32 real-runner suites (`packages/sandbox/sandbox-windows-acl/tests/`, `packages/bash/pwsh-sandbox/tests/`) are the substitute evidence, and the CI Windows lane owns the assembled signal. The grant machinery is pinned cross-platform by `packages/sandbox/sandbox-local/tests/acl-grants.spec.ts` (the derived private-temp identity — deterministic per session + workspace, distinct across sessions — one-shot materialization, exclusive temp creation with reparse-point rejection and self-cleanup on failure, clean-restart re-grant of the same derived directory, the standing-vs-revocable lifecycle across dispose and the mode-switch cycle, and the derived-SID argv contract — with the Win32 surface mocked) and on win32 by `workspace-sid.spec.ts` (derivation determinism/shape/distinctness), `grant.spec.ts` (real-DACL materialization: revocable paths revoke on dispose, standing paths survive it), the `acl.spec.ts` idempotent-grant fast-path and standing-ACE-after-dispose contract, the `failure-paths.spec.ts` suspension-orphan regression (AssignProcessToJobObject failure terminates the child), and the `runner.spec.ts` `--write-sid` contract (caller-owned grants, private temp subdir through TMP/TEMP, both-mode CIM-denial probes, the mode-downgrade regression — a standing grant ACE is inert under read-only and effective again on re-upgrade — the ambient-writable Public-probe regression (a C:\Users\Public subdirectory write is denied under both modes), and the ConstrainedLanguage pins in both modes, plus the grandchild-stdio matrix pins — inherited/ignored stdio spawns succeed while piped capture is DENIED in both modes). The runner-failure classification is exit-gated on 127 (a confined command that merely prints the `windows-acl-run:` signature on a non-127 exit is never misclassified as "the command did not run" — pinned in the pwsh-sandbox helper suite).
|
||||
The product-visible Windows roster flip is win32-only, so keyless snapshots that must replay on macOS/Linux cannot cover it; bundle composition specs plus the win32 real-runner suites are the substitute evidence, and the CI Windows lane owns the assembled signal. `sandbox-local/tests/acl-grants.spec.ts` pins random temp allocation, per-session/workspace reuse, fork/workspace separation, crash-resume non-collision, paired argv SIDs, failure cleanup, and standing-versus-revocable lifecycle with Win32 mocked. On Windows, `workspace-sid.spec.ts` pins workspace/temp derivation and domain separation; `acl.spec.ts` pins real DACL lifecycle; and `runner.spec.ts` pins paired-SID validation, sibling temp denial under a shared workspace SID, per-call agentless temp creation/removal, TMP/TEMP rewriting, mode downgrade, Public denial, Everyone/hard-link partial boundaries, mode-specific PowerShell language behavior, and grandchild stdio. ARM64 and emulated x64 native runs own the architecture-specific acceptance evidence.
|
||||
|
||||
## Related
|
||||
|
||||
|
||||
@@ -6,15 +6,15 @@ Status: implemented
|
||||
|
||||
## Problem
|
||||
|
||||
[沙盒决策](2026-07-06-sandbox.md)把 `PLATFORM_CHAINS.win32` 留空，交付的 Windows profile 因为没有可用的隔离执行器而退化为 danger-full-access。win32 档必须实现沙盒词汇表承诺的两个文件效果模式——`read-only`（零写入）与 `workspace-write`（仅工作区根目录加后端定义的临时区域可写）——同时保持读、网络与进程可见性不受影响，因为所有模式都允许读取。
|
||||
最初的[沙箱决策](2026-07-06-sandbox.md)将 `PLATFORM_CHAINS.win32` 留空，因此交付的 Windows profile 因不存在隔离执行器而退化为 danger-full-access。win32 档必须约束沙箱词汇表中的两种文件效果模式——`read-only`（不显式授予任何可写根目录）与 `workspace-write`（允许写入工作区根目录及后端定义的临时区域）——并报告其机制无法约束的任何效果；读取、网络与进程可见性仍在这套词汇之外。
|
||||
|
||||
## Decision
|
||||
|
||||
直接基于原始 ACL 机制实现该档：把调用者令牌复制为 `WRITE_RESTRICTED` 受限令牌（`CreateRestrictedToken`，`WRITE_RESTRICTED` + `DISABLE_MAX_PRIVILEGE` + `LUA_TOKEN`），其 restricting SIDs 中包含写入 SID（`S-1-4-x-y`）；工作区与临时目录上写入 SID 的 Write ACE 就是全部写入白名单，因为 `WRITE_RESTRICTED` 只对写访问做交集检查，读保持调用者的完整环境访问。该机制来自 huoyaoyuan/windows-acl-restrict-poc（`10e4dfb`）的演示；本移植检查每一个 API 调用并 fail-closed（POC 因忽略返回值而 fail-open）。写入 SID 是**按工作区**的身份，由规范工作区路径确定性派生（`workspaceWriteSid`——sha256 → `S-1-4-x-y`），且**任何地方都不存储**：工作区根目录 ACE 因此每台机器每个工作区只物化一次——常驻 ACE 就是跨会话复用缓存，此后每次供给都命中精确 ACE 跳过（幂等重授权跳过急切的全树重传播——不做垃圾回收）——而不是每会话一次，这正是先前每会话随机 SID 每个会话都要付一次全树传播的代价。seam 从会话 id + 工作区派生会话的**私有**临时子目录（sha256、16 位 hex——任何地方都不存储，因此不存在篡改面）并独占创建；它在提供方 dispose 时移除，崩溃则把它留作 `%TEMP%` 垃圾，其下一次恢复会在独占创建处大声失败，直到临时目录卫生机制将其回收。seam 把工作区 ACE **常驻**物化（绝不撤销——就是缓存），把临时 ACE **可回收**物化（提供方 dispose（资源释放）时撤销，因此可继承 ACE 不会在环境临时根目录上比其会话的临时目录活得更久）。令牌的 restricting list 是保活组加上仅 workspace-write 下的写入 SID：read-only = [登录 SID、Everyone]，workspace-write = [登录 SID、Everyone、写入 SID]。保活不变式是登录 SID + Everyone（没有它们，早期 DLL init 会以 0xC0000142 死亡，CNG 会让 pwsh 以 0xE0434352 崩溃）。Read-only 不含写入 SID：先前 workspace-write 时期留下的常驻授权 ACE 保持**失效**（pass-2 检查只授予列表所携带的内容，因此 read-only 在 `/permission` 降级或崩溃后恢复的会话中始终保持严格零授权，而常驻 ACE 让重新升级保持零成本）。Authenticated Users 在**两种**列表中都缺席——WMI namespace 安全校验失败（0x80041003），因此 CIM 在每一种受限模式下都不可用，且 C:\-root 建树逃逸（驻留的 `AU:(AD)` + `AU:(OI)(CI)(IO)(M)` ACE）在两种模式下都被关闭；INTERACTIVE/LOCAL 同样在两种列表中都缺席（Public 树的写入被拒绝——由 runner 的 Public-probe 回归钉住）。Workspace-write 子进程看到的是私有的每会话临时子目录（`<temp>\dsh-<16 hex>`——由会话 id + 工作区派生、独占创建、拒绝 reparse point、提供方 dispose 时移除——TMP/TEMP 由 runner 重写——bwrap `--tmpfs /tmp` 语义）。受限令牌的**默认 DACL** 被扩展一条写入 SID 全权 ACE（`SetTokenInformation(TokenDefaultDacl)`）：此后不带显式安全描述符创建的新对象（匿名管道——CreatePipe、同步对象）自带 restricting SID ACE，创建时的写 pass-2 检查通过；**named pipe 例外**——其默认安全描述符是 Win32 层在用户态安装的默认 SD 模板（由 KernelBase 构建——owner/SYSTEM/Admins 全权、Everyone/ANONYMOUS 只读），令牌无法影响，因此受限孙进程的管道 stdio 捕获保持拒绝（POC 记载的边界，由 runner 套件钉住）。它以 [`@deepseek-ai/dsh-sandbox-windows-acl`](../../../../packages/sandbox/sandbox-windows-acl/README.md)（后端加 `./runner` argv 前缀入口）、[`dsh-sandbox-local`](../../../../packages/sandbox/sandbox-local/README.md) 的 `win32` 链档、以及作为隔离执行器的 [`@deepseek-ai/dsh-pwsh-sandbox`](../../../../packages/bash/pwsh-sandbox/README.md) 交付；Windows 平台层在受限 pwsh 栈之上重新启用完整权限面（sandbox/sandbox-policy/permission/approval/fs-sandbox）。
|
||||
直接基于原始 ACL 机制实现该档：把调用者令牌复制为 `WRITE_RESTRICTED` 受限令牌（`CreateRestrictedToken`，`WRITE_RESTRICTED` + `DISABLE_MAX_PRIVILEGE` + `LUA_TOKEN`），其 restricting SIDs 携带彼此独立的工作区能力与私有临时目录能力。`WRITE_RESTRICTED` 只对写访问做交集检查，因此读取保留调用者的环境访问，而写入还必须匹配这些能力 ACE 之一。该机制来自 huoyaoyuan/windows-acl-restrict-poc（`10e4dfb`）的演示；本移植检查每个 API 调用并 fail-closed（POC 因忽略失败而 fail-open）。工作区 SID 由规范工作区路径确定性派生（`workspaceWriteSid`——sha256 → `S-1-4-x-y`）；其常驻工作区 ACE 是跨会话复用缓存，精确 ACE 跳过可避免重复的急切全树传播。每个活跃的会话/工作区对则获得一个随机私有临时目录，以及一个从该路径派生的、经过域分离的 SID（`tempWriteSid`）；其 ACE 可回收，TMP/TEMP 指向该目录，令牌默认 DACL 列入该临时 SID，因此新建的临时对象不会获得共享的工作区能力。fork 因此无法写入同级会话的临时目录树。即使恢复的是同一会话，新的提供方也会选择新的路径和 SID，因此崩溃残留只是失效垃圾，而非冲突或继承的能力；无 agent（智能体）的调用会逐调用创建并移除同样的形态。环境临时根目录绝不会被隐式授权。如果工作区等于或包含临时根目录，调用会在任何 ACL 改动发生前失败，因为否则其可继承的常驻 ACE 会向每个私有子目录授权；直接 API 会拒绝可写根目录与实际私有临时目录在任一方向上的重叠。PowerShell 可借助这项私有临时目录能力完成启动时的 AppLocker 探针，因此在没有主机范围策略时，`workspace-write` 会保持 FullLanguage；`read-only` 无法创建探针文件，会保守地进入 ConstrainedLanguage。这一区别属于 PowerShell 启动行为，不是 ACL 边界的一部分。令牌列表为 read-only = [登录 SID、Everyone]，workspace-write = [登录 SID、Everyone、工作区 SID、可选临时 SID]。登录 SID + Everyone 是保活不变式（没有它们，早期 DLL 初始化会以 0xC0000142 死亡，CNG 会让 pwsh 以 0xE0434352 崩溃）。由于 Everyone 仍在列表中，向 Everyone 授予写访问的外部对象会通过两次检查；由于 NTFS ACL 属于文件对象，工作区内获授权的硬链接也会使同一对象的外部别名获得授权。拒绝所有硬链接会让普通 pnpm 工作区不可用，因此提供方报告 `enforcement: 'partial'`，原生套件则钉住这两个缺口。Read-only 不含任何能力 SID，因此常驻工作区 ACE 在模式降级后保持失效。Authenticated Users 在两种列表中都不存在——CIM 不可用，从而关闭 C:\-root 建树逃逸——INTERACTIVE/LOCAL 也不存在，因此 Public 树写入被拒绝。新建匿名管道和同步对象通过 `SetTokenInformation(TokenDefaultDacl)` 继承临时 SID（禁用临时目录时继承工作区 SID，read-only 下继承 Everyone）；named pipe 保持 Win32 层 owner/SYSTEM/Admins 全权、Everyone/ANONYMOUS 只读的模板，因此受限孙进程的管道 stdio 仍被拒绝。它以 [`@deepseek-ai/dsh-sandbox-windows-acl`](../../../../packages/sandbox/sandbox-windows-acl/README.md)、[`dsh-sandbox-local`](../../../../packages/sandbox/sandbox-local/README.md) 的 `win32` 档，以及作为隔离执行器的 [`@deepseek-ai/dsh-pwsh-sandbox`](../../../../packages/bash/pwsh-sandbox/README.md) 交付。
|
||||
|
||||
## How the restriction works (why no new identity)
|
||||
|
||||
身份路线靠"**谁**在跑子进程"来限制，本档靠"令牌派生"来限制。身份路线（landstrip 的 restricted-user、AppContainer）用全新账户或容器 SID 运行子进程，该身份在宿主的文件上从零条 ACE 开始——一切访问（包括读）默认拒绝，子进程要碰的每条路径都必须事后为那个身份补写 ACE 才能放行：这正是让两个备选方案出局的全盘 DACL 改造。受限令牌保留调用者自己的 SID 与 logon session：[`CreateRestrictedToken`](https://learn.microsoft.com/en-us/windows/win32/api/securitybaseapi/nf-securitybaseapi-createrestrictedtoken) 派生一个加入 restricting SIDs 与 `WRITE_RESTRICTED` 标志的令牌，于是 Windows 做两次访问检查——一次按正常 SID，一次按 restricting SIDs——只有两次都放行，写类访问才被授予。读只凭正常检查即可通过（调用者的 SID 在其可读范围内本来就携带读权限），所以本档不需要任何读授权、也不需要新账户；写还必须额外通过孤儿 SID 检查，而只有工作区与临时目录的 ACE 能满足它。`DISABLE_MAX_PRIVILEGE | LUA_TOKEN` 在令牌侧合成了新账户的受限用户效果，即使提升过的调用者派生的也是过滤令牌。同一原语其实也能限制读（`SidsToDisable` 把 SID 变为 deny-only），但受限读的令牌需要逐路径的读授权——恰好重新引入身份路线付出的代价——而沙盒词汇表从不要求读隔离。
|
||||
身份路线靠"**谁**在跑子进程"来限制，本档靠"令牌派生"来限制。身份路线（landstrip 的 restricted-user、AppContainer）用全新账户或容器 SID 运行子进程，该身份在宿主的文件上从零条 ACE 开始——一切访问（包括读）默认拒绝，子进程要碰的每条路径都必须事后为那个身份补写 ACE 才能放行：这正是让两个备选方案出局的全盘 DACL 改造。受限令牌保留调用者自己的 SID 与 logon session：[`CreateRestrictedToken`](https://learn.microsoft.com/en-us/windows/win32/api/securitybaseapi/nf-securitybaseapi-createrestrictedtoken) 派生一个加入 restricting SIDs 与 `WRITE_RESTRICTED` 标志的令牌，于是 Windows 做两次访问检查——一次按正常 SID，一次按 restricting SIDs——只有两次都放行，写类访问才被授予。读只凭正常检查即可通过（调用者的 SID 在其可读范围内本来就携带读权限），所以本档不需要任何读授权、也不需要新账户；写还必须额外通过能力 SID 检查，而只有工作区与临时目录的 ACE 能满足它。`DISABLE_MAX_PRIVILEGE | LUA_TOKEN` 在令牌侧合成了新账户的受限用户效果，即使提升过的调用者派生的也是过滤令牌。同一原语其实也能限制读（`SidsToDisable` 把 SID 变为 deny-only），但受限读的令牌需要逐路径的读授权——恰好重新引入身份路线付出的代价——而沙盒词汇表从不要求读隔离。
|
||||
|
||||
## Alternatives considered
|
||||
|
||||
@@ -32,11 +32,11 @@ AppContainer 令牌没有环境读访问：每个可读路径都必须预先通
|
||||
|
||||
## Consequences
|
||||
|
||||
所得：仅写隔离、不引入新的 OS 版本下限（`CreateRestrictedToken` 比 mxc 的版本早二十年）、读/网络/进程可见性完全不受影响（与模式词汇表一致）、fail-closed 错误携带 API 名与精确 Win32 错误码。所失：无读侧或网络隔离；控制台隔离不可用（隐藏控制台子进程以 `STATUS_DLL_INIT_FAILED` 死亡；子进程共享宿主控制台）；被授权根目录上有驻留 ACE 改动（目录须为调用者所有；工作区 ACE 按设计永久常驻——复用缓存，工作区改名时成为不可见残留——临时 ACE 由提供方 dispose 连同派生的私有临时目录一起回收——崩溃会把两者都留下，下一次恢复会在独占创建处大声失败，直到临时目录卫生回收该目录）；授权物化是急切的全树传播（`SetNamedSecurityInfoW` 立即遍历每个后代——在大型工作区上耗时数十秒），因按工作区身份，每台机器每个工作区只付一次；CIM 在**两种**受限模式下都不可用（AuthUsers 从两种列表中被移除——WMI namespace 安全校验失败，`Get-ComputerInfo` 静默返回不完整结果），作为关闭两种模式下 C:\-root 建树逃逸的代价；位于被授权根目录之外的 FAT 类（无 ACL）目标在两种模式下仍可写（没有可做交集的安全描述符——作为历史残留处理：不支持、仅警告、已在 README 中记录）；NULL DACL 目录在 grant+revoke 往返下不保持身份（记录在案的边角，POC 亦有此行为）；`whoami` 与令牌检查 cmdlet 在受限令牌下失败（诊断噪音，已记录）；且**两种**受限模式都以 ConstrainedLanguage 模式运行 `pwsh`——受限令牌触发 PowerShell 的锁定检测，因此 `Add-Type`、非核心 .NET 静态调用（`[System.IO.*]::`、`[math]::`）、COM 对象与反射都会以“only core types”错误失败，而 `-f` 格式化、属性访问与核心 cmdlet/类型继续工作，语言模式也无法从内部提升回 FullLanguage——已在 pwsh 工具描述中教给模型，并记录在包 README 的 Known Limitations 中；**两种**受限模式同样拒绝 named-pipe 打开——libuv 的管道 stdio spawn 以 EPERM 失败（POC 记载的“无法重定向输出”边界；继承/忽略的 stdio 与匿名管道可用）——记录在包 README 的 Known Limitations 中，并在 pwsh 工具描述中教给模型。
|
||||
所得：仅写隔离、不引入新的 OS 版本下限（`CreateRestrictedToken` 比 mxc 的版本早二十年）、读/网络/进程可见性完全不受影响（与模式词汇表一致），且 fail-closed 错误携带 API 名与精确 Win32 错误码。会话共享有意常驻的工作区能力，但不共享各自可回收的临时能力；重启残留既不能阻塞恢复的会话，也不能向其授权。所失：强制执行在结构上只能是部分的，因为此令牌形态无法把 Everyone 授予的写入与 NTFS 硬链接别名限制在路径边界内；无读侧或网络隔离；控制台隔离不可用（隐藏控制台子进程以 `STATUS_DLL_INIT_FAILED` 死亡；子进程共享宿主控制台）；工作区常驻 ACE 改动（复用缓存，以及工作区改名后的失效残留）与异常关闭后遗留的随机临时目录垃圾，直到 OS 卫生机制将其回收；工作区授权采用急切的全树传播（`SetNamedSecurityInfoW` 立即遍历每个后代——大型工作区上耗时数十秒），每台机器每个工作区只付一次；CIM 在两种受限模式下均不可用（Authenticated Users 不存在，从而关闭 C:\-root 建树逃逸）；FAT 类无 ACL 目标仍可写；NULL-DACL 目录在 grant/revoke 往返下不保持身份；`whoami` 与令牌检查 cmdlet 在受限令牌下失败；read-only pwsh 会进入 ConstrainedLanguage，而在没有主机策略时 workspace-write 保持 FullLanguage；named pipe 打开仍被拒绝，因此 libuv 管道 stdio 的孙进程以 EPERM 失败，而继承/忽略的 stdio 与匿名管道可用。包 README 负责记录这些运行限制。
|
||||
|
||||
## Testing
|
||||
|
||||
产品可见的 Windows 阵容切换仅存在于 win32，而 keyless 快照夹具必须在 macOS/Linux 上可重放，因此无法覆盖它；替代证据是 bundle 组合 spec（[`base.spec.ts`](../../../../packages/bundle/base/tests/base.spec.ts)、[`windows-shell.spec.ts`](../../../../apps/cli/tests/windows-shell.spec.ts)）加上 win32 真实 runner 套件（`packages/sandbox/sandbox-windows-acl/tests/`、`packages/bash/pwsh-sandbox/tests/`），组装态信号由 CI 的 Windows lane 负责。授权机制在跨平台侧由 `packages/sandbox/sandbox-local/tests/acl-grants.spec.ts` 钉住（派生的私有临时身份——按会话 + 工作区确定性、跨会话相异——一次性物化、独占临时目录创建并拒绝 reparse point、失败时自我清理、干净重启时对同一派生目录的重新授权、dispose 与模式切换循环中的常驻/可回收生命周期，以及派生 SID 的 argv 契约——mock 掉 Win32 表面），win32 侧由 `workspace-sid.spec.ts`（派生的确定性/形态/相异性）、`grant.spec.ts`（真实 DACL 物化：可回收路径在 dispose 时撤销、常驻路径存活）、`acl.spec.ts` 的幂等授权快速路径与 dispose 后常驻 ACE 契约、`failure-paths.spec.ts` 的 suspension-orphan 回归（AssignProcessToJobObject 失败会终止子进程）与 `runner.spec.ts` 的 `--write-sid` 契约（调用者所有目录的授权、经 TMP/TEMP 的私有临时子目录、两种模式下的 CIM 拒绝探针、模式降级回归——驻留授权 ACE 在 read-only 下失效并在重新升级后再度生效——环境可写 Public-probe 回归（对 C:\Users\Public 子目录的写入在两种模式下都会被拒绝），以及两种模式下对 ConstrainedLanguage 的钉定，加上孙进程 stdio 矩阵钉定——继承/忽略的 stdio spawn 成功，而管道捕获在两种模式下都被**拒绝**）钉住。runner 失败分类以 127 退出码为门槛（受限命令仅仅在非 127 退出时打印 `windows-acl-run:` 签名，也绝不会被误分类为"命令未运行"——由 pwsh-sandbox helper 套件钉住）。
|
||||
产品可见的 Windows 阵容切换仅存在于 win32，而必须在 macOS/Linux 上可重放的 keyless 快照无法覆盖它；替代证据是 bundle 组合 spec 加上 win32 真实 runner 套件，组装态信号由 CI 的 Windows lane 负责。`sandbox-local/tests/acl-grants.spec.ts` 在 mock Win32 的情况下钉住随机临时目录分配、按会话/工作区复用、fork/工作区分离、崩溃后恢复不冲突、成对 argv SID、失败清理，以及常驻/可回收生命周期。在 Windows 上，`workspace-sid.spec.ts` 钉住工作区/临时目录派生与域分离；`acl.spec.ts` 钉住真实 DACL 生命周期；`runner.spec.ts` 钉住成对 SID 验证、共享工作区 SID 时对同级会话临时目录的拒绝、无 agent 调用的逐调用临时目录创建/移除、TMP/TEMP 重写、模式降级、Public 拒绝、Everyone/硬链接部分边界、按模式区分的 PowerShell 语言行为与孙进程 stdio。ARM64 与模拟 x64 原生运行负责提供架构特定的验收证据。
|
||||
|
||||
## Related
|
||||
|
||||
|
||||
@@ -0,0 +1,6 @@
|
||||
# Bilingual-pair consistency record (docs/i18n/README.md): the git blob hash of each
|
||||
# side as of the last confirmed-consistent state. Both languages carry equal authority;
|
||||
# after editing either side, bring the other along and re-record with:
|
||||
# pnpm run verify-translation-pairing --write .agents/notes/implemented/feature/2026-08-10-web-session-log-export.md
|
||||
2026-08-10-web-session-log-export.md: 427b6478ac44fb28030aa932630f276de7bb2edc
|
||||
2026-08-10-web-session-log-export.zh.md: 63b9804a54cda7eea4ff793d78a925fe296d06cb
|
||||
@@ -0,0 +1,30 @@
|
||||
# Agent Note: Web session-log export as a host-streamed ZIP download
|
||||
|
||||
Status: implemented
|
||||
|
||||
English | [中文](2026-08-10-web-session-log-export.zh.md)
|
||||
|
||||
## Problem
|
||||
|
||||
The Trajectory view had no way to hand a debugging artifact to a human: the raw session log lived on disk and in the host, the client history face served folded projections (not raw entries), and a session with subagents spans many independent session logs. A bug report needs the complete raw log of the whole tree, in a shape that survives being emailed around.
|
||||
|
||||
## Decision
|
||||
|
||||
- **The export is a host-only download, not an RPC**: `GET /api/session.export?sessionId=…&includeDescendants=true` streams one ZIP attachment. Every file is a session's **stored artifact text verbatim**: `readRaw` on the persistence service reads the backend's own durable bytes (the JSONL backend decodes its physical zstd frames, or returns plaintext) — never a reconstruction from parsed events, so packed-chunk rows, key order, and line breaks survive byte-for-byte — under its original base name (`session.jsonl` at the root, `subagents/<id>/session.jsonl` for descendants). Compression runs on the host with fflate's streaming `Zip`/`ZipDeflate` API, each entry deflated in bounded chunks as it is produced, so the response is chunked as it is generated and the host never holds the whole archive in one buffer (at most one descendant's artifact text beyond the preloaded root), and production yields whenever the response queue fills, so a slow consumer bounds the accumulation (fflate's callback is synchronous — the drain point is the only backpressure). No manifest is written — every file is byte-identical to the durable artifact and self-describing through its own header line.
|
||||
- **Error vocabulary is HTTP-native**: missing services → 500, missing root session → 404 (both decided before any byte streams), a descendant without a stored artifact → the stream errors (fail-loud, never silent under-export). The carrier (`toFetchHandler`) already applies the `/api` trust fence; the GET branch sits beside the existing SSE GET routes, and `ApiProxy.downloads.sessionLog` (host-only, no wire envelope, absent from `IApiClient`) implements it.
|
||||
- **The UI just downloads**: the 导出 button fetches the endpoint and saves the response; the `session.log` RPC that an earlier iteration shipped was removed — the download endpoint is its only consumer, and the repo rule is no public interface without a current owner. The client bundle no longer carries fflate (the earlier browser-entry-alias pitfall is moot).
|
||||
- The 导出 button lives in the Trajectory toolbar; the plugin exposes `exportLog` through the view's inject face (components never touch ctx) and resolves the view tab label through the locale service (`轨迹` in Chinese, `Trajectory` in English). In-flight state disables the button; a failure surfaces in a visible alert bar under the toolbar.
|
||||
|
||||
## Alternatives considered
|
||||
|
||||
- **`session.log` data RPC + client-side zip** — shipped first, rejected with the user: the browser pulls the full raw JSON (≈10× the final zip size) and compresses on the main thread; for the 23 MB sessions in real use the host-side stream is strictly better. The RPC was deleted with the migration rather than left as a dead public surface.
|
||||
- **Single JSONL with envelope lines for multiple sessions** — rejected with the user: mixing sessions in one JSONL loses clean per-file boundaries; a ZIP keeps one canonical file per session.
|
||||
- **jszip** — heavier (~100 kB) and its dependency graph pulls readable-stream browser mappings; fflate is purpose-built and small.
|
||||
- **Vendoring fflate's browser entry** — the repo vendoring procedure targets cordis-scale pinned sources; a resolveId alias keeps the maintained dependency without shipping a copy (and host-side fflate needs no alias at all).
|
||||
|
||||
## Consequences
|
||||
|
||||
- Export fidelity: every exported file is byte-identical to the backend's durable artifact as of the read moment (a live session may append after the read; the export reflects the durable state at read time). The archive name is `dsh-session-<sanitized-id>.zip` and archive paths sanitize ids before they can shape entries.
|
||||
- `readRaw` joins the persistence service as a concrete default (`undefined` for backends without a per-session artifact, e.g. SQLite) with a JSONL-backend override that owns the compression decode. `ApiProxy.downloads.sessionLog` adds one host-only member to the contract plus a host-side query schema and a GET branch in the fetch handler — no RPC map row, envelope schema, or client `IApiClient` surface.
|
||||
- Fixture mode (no host) answers 404 for the export, so the button's error bar explains the gap instead of hanging; the navigation-panes golden snapshot includes the 导出 button.
|
||||
- Deferred: transcript.md and a report/feedback bundle remain future work; the byte-faithful, manifest-free shape keeps the v2 bundle extension cheap.
|
||||
@@ -0,0 +1,30 @@
|
||||
# Agent Note：Web 会话日志导出——宿主流式 ZIP 下载
|
||||
|
||||
状态：implemented
|
||||
|
||||
[English](2026-08-10-web-session-log-export.md) | 中文
|
||||
|
||||
## 问题
|
||||
|
||||
Trajectory 视图没有任何方式把调试工件交到人手里：原始会话日志存放在磁盘与宿主侧，客户端历史面只提供折叠后的投影（而非原始事件），而带子代理的会话横跨多个相互独立的会话日志。bug 报告需要整棵会话树的完整原始日志，并且形态要能在被转发后仍然可用。
|
||||
|
||||
## 决策
|
||||
|
||||
- **导出是宿主侧的下载面，不是 RPC**：`GET /api/session.export?sessionId=…&includeDescendants=true` 流式返回一个 ZIP 附件。每个文件都是会话**存储工件的逐字原文**：持久化服务新增的 `readRaw` 读取后端自己的持久化字节（jsonl 后端解码其物理 zstd 帧，或直接返回明文）——绝非从解析后事件重建，因此 chunk 打包、键序、换行全部逐字节保留——放在其原始基础文件名下（根为 `session.jsonl`，子代理为 `subagents/<id>/session.jsonl`）。压缩在宿主侧用 fflate 的流式 `Zip`/`ZipDeflate` API 完成，每个条目按有界分块边产出边压缩，响应随生成分块写出，宿主从不把整个归档放进单个缓冲区（除预载的根外，最多同时持有一条后代的工件文本），且每当响应队列填满时生产会让出，慢消费者因此只产生有界的积压（fflate 的回调是同步的——让出点是唯一的背压手段）。不写清单——每个文件都与持久化工件逐字节一致，并通过自身 header 行自描述。
|
||||
- **错误词汇是 HTTP 原生的**：服务缺失 → 500，根会话缺失 → 404（两者都在任何字节流出前判定），后代缺少存储工件 → 流失败（fail-loud，绝不静默少导出）。载体（`toFetchHandler`）已对 `/api` 应用信任围栏；GET 分支与既有 SSE GET 路由并列，由 `ApiProxy.downloads.sessionLog`（host-only、无 wire 信封、不在 `IApiClient` 上）实现。
|
||||
- **UI 只负责下载**：「导出」按钮 fetch 该端点并保存响应；早先迭代发布的 `session.log` RPC 已删除——下载端点是它唯一的消费者，仓库规则是不留无当前所有者的公共接口。客户端 bundle 不再携带 fflate（早先的浏览器入口别名坑随之消失）。
|
||||
- 「导出」按钮位于 Trajectory 工具栏；插件通过视图的 inject face 暴露 `exportLog`（组件从不接触 ctx），并通过 locale 服务解析视图标签页标题（中文「轨迹」、英文 "Trajectory"）。进行中状态会禁用按钮；失败会在工具栏下方的可见警示条中显示。
|
||||
|
||||
## 考虑过的替代方案
|
||||
|
||||
- **`session.log` 数据 RPC + 客户端打包**——先发布，后与用户共同否决：浏览器要拉取完整原始 JSON（约为最终 zip 的 10 倍）并在主线程压缩；对实际使用中 23 MB 级别的会话，宿主流式严格更优。迁移时把该 RPC 一并删除，而不是留作无消费者的公共接口。
|
||||
- **用信封行把多会话编码进单一 JSONL**——与用户共同否决：把多个会话混进一个 JSONL 会失去干净的按文件边界；ZIP 让每个会话保持一个规范文件。
|
||||
- **jszip**——更重（约 100 kB），依赖图还会拉入 readable-stream 的浏览器映射；fflate 专为此而生且体积小。
|
||||
- **将 fflate 浏览器入口 vendoring 进仓库**——仓库的 vendoring 流程面向 cordis 级别的固定源码；resolveId 别名在保持维护中的依赖的同时无需复制代码（宿主侧 fflate 根本不需要别名）。
|
||||
|
||||
## 后果
|
||||
|
||||
- 导出保真度：每个导出文件都与读取时刻的后端持久化工件逐字节一致（活跃会话可能在读取后继续追加；导出反映的是读取时的持久化状态）。压缩包名为 `dsh-session-<sanitized-id>.zip`，归档路径在塑造条目前会先净化会话 id。
|
||||
- `readRaw` 以具体默认（无每会话工件的后端如 SQLite 返回 `undefined`）加入持久化服务，jsonl 后端覆写并自持压缩解码。`ApiProxy.downloads.sessionLog` 为契约新增一个 host-only 成员，外加宿主侧 query schema，并在 fetch handler 加一个 GET 分支——没有 RPC map 行、信封 schema 或客户端 `IApiClient` 面。
|
||||
- fixture 模式（无宿主）对导出应答 404，按钮的错误条会解释这个缺口而非挂起；navigation-panes golden 快照包含「导出」按钮。
|
||||
- 暂缓：transcript.md 以及 report/feedback 打包留待后续；逐字节忠实、无清单的形态让 v2 的打包扩展保持廉价。
|
||||
Reference in New Issue
Block a user