Merge pull request #962 from deepseek-harness/codex/sandbox-policy-context
feat(sandbox-policy): expose current file policy to models
This commit is contained in:
Tianyi Cui
@@ -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: efb4159d736779af28edc1ae6091de4669c92f31
|
||||
2026-07-06-approval-seam.zh.md: 9a4656f30a43473fe90cde9c56d45f48943e5d10
|
||||
2026-07-06-approval-seam.md: ae143a41302b7bcd6029345fa91ca4eda837c141
|
||||
2026-07-06-approval-seam.zh.md: a66ce804167fd4de5c8dbe51d70c9b038d1ebb17
|
||||
|
||||
@@ -25,7 +25,7 @@ One `cordis.yml` entry mounts the seam. Not loading it is the fail-closed opt-ou
|
||||
# policy: never # deployment default for sessions without an override; 'ask' when omitted
|
||||
```
|
||||
|
||||
The entry alone provides mechanism, not a channel: with no answerer composed, every ask resolves `unavailable` and the asking tool call denies — fail-closed needs no configuration. Composing the ACP app (`@deepseek-ai/dsh-acp-demo`, as in [the acp-agent example's default tree](../../../../examples/acp-agent/README.md)) completes the loop: its [automation-only bridge](../simplification/2026-07-23-acp-automation-only-protocol.md) registers an answerer that sends `session/request_permission` to the owning client with the exact tool-call id and one-shot allow/reject options. `policy: never` is the unattended stance — every ask auto-rejects deterministically and is stated in the system prompt. `policy` is validated against the closed list at plugin load; anything else throws.
|
||||
The entry alone provides mechanism, not a channel: with no answerer composed, every ask resolves `unavailable` and the asking tool call denies — fail-closed needs no configuration. Composing the ACP app (`@deepseek-ai/dsh-acp-demo`, as in [the acp-agent example's default tree](../../../../examples/acp-agent/README.md)) completes the loop: its [automation-only bridge](../simplification/2026-07-23-acp-automation-only-protocol.md) registers an answerer that sends `session/request_permission` to the owning client with the exact tool-call id and one-shot allow/reject options. `policy: never` is the unattended stance — every ask auto-rejects deterministically, and the current value joins the runtime-context snapshot. `policy` is validated against the closed list at plugin load; anything else throws.
|
||||
|
||||
What a composed deployment observes: `allowed-once` lets exactly that call proceed; rejection, dismissal, and channel absence deny with three distinct reasons the model can tell apart; a successful in-turn request lands a durable `approval/asked`/`approval/decided` pair on the asking agent's session log; nothing about a grant persists past the call that asked. An idle request or audit append failure rejects instead of returning an unaudited decision.
|
||||
|
||||
@@ -63,7 +63,7 @@ Answerers are `approval/request` waterfall listeners. Zero listeners fall throug
|
||||
|
||||
#### The per-session policy tier
|
||||
|
||||
The seam also owns the session-scoped `'ask' | 'never'` policy described by [the sandbox Agent Note](2026-07-06-sandbox.md). Effective policy is folded from logged switches over the deployment default. `'never'` resolves to `rejected` inside `request()` before any answerer can run; `'ask'` dispatches and otherwise falls through to `unavailable`. The prompt states only deterministic `'never'`, switch narration is coalesced, and every request still records the audit pair.
|
||||
The seam also owns the session-scoped `'ask' | 'never'` policy described by [the sandbox Agent Note](2026-07-06-sandbox.md). Effective policy is folded from logged switches over the deployment default. `'never'` resolves to `rejected` inside `request()` before any answerer can run; `'ask'` dispatches and otherwise falls through to `unavailable`. Both current values join the atomic runtime-context snapshot before each model request, so a policy switch needs no separate narration; every approval request still records the audit pair.
|
||||
|
||||
#### The ACP answerer
|
||||
|
||||
@@ -83,7 +83,7 @@ The answerer routes through the bridge's exact-agent ownership check described b
|
||||
|
||||
Unit tests pin outcomes, first-wins delegation, containment, cancellation, scoped routing, audit pairing, the unbypassable `'never'` policy, tool deny reasons, and ACP ownership/outcome mapping through a real scripted bridge.
|
||||
|
||||
Snapshots record allowed and rejected sandbox escalation through `session/request_permission`, plus the `'never'` prompt and policy-switch notice. Unscripted permission prompts cancel and fail closed.
|
||||
Snapshots record allowed and rejected sandbox escalation through `session/request_permission`, plus the complete `'ask'` and `'never'` runtime-context contributions. Unscripted permission prompts cancel and fail closed.
|
||||
|
||||
## Deferred
|
||||
|
||||
@@ -124,7 +124,7 @@ Costs and accepted limits:
|
||||
- **What happens when the user dismisses the prompt, or the turn aborts mid-ask?** Dismissal maps to `cancelled` with its own deny text. An already-aborted signal settles `cancelled` without dispatching; an abort during the ask discards the late answer. When both audit appends commit, either path records one pair, never two.
|
||||
- **What if the client answers with an option the harness never offered?** Any selection other than the offered `allow_once` maps to `rejected` — an unknown optionId from a non-conforming client can never grant.
|
||||
- **How do subagents' approvals route?** An agent no answerer owns delegates through the whole waterfall and fails closed — in-process subagents are deliberately unanswerable. A `'never'` parent seeds that override into each in-process child's log ([decision](2026-07-25-subagent-policy-inheritance.md)), so the child is told up front instead of asking into the empty waterfall. `subagent-acp`'s child-side auto-answer is separate; routing a child's asks to the parent controller is deferred (§ Deferred).
|
||||
- **What does `policy: 'never'` actually change at runtime?** The service resolves every ask for that session to `rejected` before dispatching any answerer (in-service, so no registration order can bypass it); the system prompt states the policy; switches are narrated at boundaries; each successful auto-rejection records the audit pair.
|
||||
- **What does `policy: 'never'` actually change at runtime?** The service resolves every ask for that session to `rejected` before dispatching any answerer (in-service, so no registration order can bypass it); the next atomic runtime-context snapshot states the policy; each successful auto-rejection records the audit pair.
|
||||
- **What happens across a hot reload, or when an answerer unloads mid-session?** Answerers dispose with their owning fiber, so the next ask degrades to `unavailable` instead of hanging on a dead channel; remounting re-registers the answerer with no catch-up state.
|
||||
- **Where does a client get approval context?** The request carries the exact `callId` and the asker's human-readable `reason`; channel adapters may correlate richer tool-call state without duplicating arguments in the approval seam.
|
||||
|
||||
|
||||
@@ -25,7 +25,7 @@ Status: implemented
|
||||
# policy: never # deployment default for sessions without an override; 'ask' when omitted
|
||||
```
|
||||
|
||||
仅有这条条目只提供机制，不提供通道：没有组合应答者时，每次 ask 都解析为 `unavailable`，发起请求的工具调用被拒绝——失败关闭无需配置。组合 ACP 应用（`@deepseek-ai/dsh-acp-demo`，如 [acp-agent 示例的默认树](../../../../examples/acp-agent/README.md)）即可闭环：其[仅面向自动化的桥接层](../simplification/2026-07-23-acp-automation-only-protocol.md)注册一个应答者，向拥有该会话的客户端发送 `session/request_permission`，携带精确的工具调用 id 和一次性 allow/reject 选项。`policy: never` 是无人值守姿态：每次 ask 确定性地自动拒绝，并在系统提示词中声明。`policy` 在插件加载时对照封闭列表校验；非法值直接抛异常。
|
||||
仅有这条条目只提供机制，不提供通道：没有组合应答者时，每次 ask 都解析为 `unavailable`，发起请求的工具调用被拒绝——失败关闭无需配置。组合 ACP 应用（`@deepseek-ai/dsh-acp-demo`，如 [acp-agent 示例的默认树](../../../../examples/acp-agent/README.md)）即可闭环：其[仅面向自动化的桥接层](../simplification/2026-07-23-acp-automation-only-protocol.md)注册一个应答者，向拥有该会话的客户端发送 `session/request_permission`，携带精确的工具调用 id 和一次性 allow/reject 选项。`policy: never` 是无人值守姿态：每次 ask 都会被确定性地自动拒绝，当前值也会加入运行时上下文快照。`policy` 在插件加载时对照封闭列表校验；非法值直接抛异常。
|
||||
|
||||
组合部署的可观测行为：`allowed-once` 仅允许该次调用继续；拒绝、关闭和通道缺失以三种不同原因拒绝，模型可以区分；轮次内成功的请求会在发起请求的 agent 的会话日志上落一对持久的 `approval/asked`/`approval/decided` 事件；授权不会在发起请求的调用结束后继续存在。空闲时的请求或审计追加失败会拒绝，而不会返回未经审计的决策。
|
||||
|
||||
@@ -63,7 +63,7 @@ tool/result "escalated" — this one call ran under the wider mode; the gra
|
||||
|
||||
#### 每会话策略层
|
||||
|
||||
seam 还拥有[沙箱 Agent Note](2026-07-06-sandbox.md) 所描述的会话级 `'ask' | 'never'` 策略。生效策略由日志中记录的切换在部署默认值之上折叠而成。`'never'` 会在任何应答者运行之前，于 `request()` 内部解析为 `rejected`；`'ask'` 则派发请求，否则一路委派至 `unavailable`。提示词仅声明确定性的 `'never'`，切换叙述会被合并，每个请求仍记录审计对。
|
||||
seam 还拥有[沙箱 Agent Note](2026-07-06-sandbox.md) 所描述的会话级 `'ask' | 'never'` 策略。生效策略由日志中记录的切换在部署默认值之上折叠而成。`'never'` 会在任何应答者运行之前，于 `request()` 内部解析为 `rejected`；`'ask'` 则派发请求，否则一路委派至 `unavailable`。两个当前值都会在每次模型请求前加入原子化的运行时上下文快照，因此策略切换无需单独叙述；每次批准请求仍会记录审计对。
|
||||
|
||||
#### ACP 应答者
|
||||
|
||||
@@ -83,7 +83,7 @@ ACP 桥只应答其会话映射所拥有的精确 agent 对象。它携带既有
|
||||
|
||||
单元测试固定结果、先到先得的委派、错误容纳、取消、作用域路由、审计配对、不可绕过的 `'never'` 策略、工具拒绝原因，以及通过真实脚本化桥实现的 ACP 归属/结果映射。
|
||||
|
||||
快照记录通过 `session/request_permission` 批准和拒绝沙箱升级，以及 `'never'` 提示词与策略切换通知。没有脚本化应答的权限提示会取消并失败关闭。
|
||||
快照记录通过 `session/request_permission` 批准和拒绝沙箱升级，以及完整的 `'ask'` 与 `'never'` 运行时上下文贡献。没有脚本化应答的权限提示会取消并失败关闭。
|
||||
|
||||
## 延后
|
||||
|
||||
@@ -124,7 +124,7 @@ ACP 桥只应答其会话映射所拥有的精确 agent 对象。它携带既有
|
||||
- **用户关闭提示或轮次在 ask 进行中中止时会发生什么？** 关闭映射为 `cancelled` 并携带自己的拒绝文本。已中止的 signal 直接结算为 `cancelled` 而不派发；ask 进行中的中止丢弃迟到的应答。当两个审计追加都提交时，任一路径都记录恰好一对事件，绝不会两对。
|
||||
- **如果客户端以 harness 从未提供的选项应答呢？** 除已提供的 `allow_once` 之外的任何选项都映射为 `rejected`——来自不合规客户端的未知 optionId 永远不能授权。
|
||||
- **subagent 的审批如何路由？** 没有应答者拥有的 agent 穿过整个 waterfall 委派并失败关闭——进程内 subagent 被刻意设计为不可应答。`'never'` 父级会把该覆盖项预置到每个进程内子 agent 的日志中（[决策](2026-07-25-subagent-policy-inheritance.md)），因此子 agent 一开始就会得知，而不是向空的 waterfall 发出 ask。`subagent-acp` 的子侧自动应答是独立的；将子 agent 的 ask 路由到父控制器已延后（§ 延后）。
|
||||
- **`policy: 'never'` 在运行时实际改变了什么？** 服务在派发任何应答者之前，将该会话的每次 ask 解析为 `rejected`（在服务内部，因此没有注册顺序能绕过它）；系统提示词声明该策略；切换在边界处被叙述；每次成功的自动拒绝都会记录审计对。
|
||||
- **`policy: 'never'` 在运行时实际改变了什么？** 服务在派发任何应答者之前，将该会话的每次 ask 解析为 `rejected`（在服务内部，因此没有注册顺序能绕过它）；下一份原子化的运行时上下文快照会声明该策略；每次成功的自动拒绝都会记录审计对。
|
||||
- **热重载或应答者在会话中途卸载时会发生什么？** 应答者随其拥有的 fiber 一起 dispose，因此下一次 ask 降级为 `unavailable` 而非挂在死通道上；重新挂载会重新注册应答者，无需追赶状态。
|
||||
- **客户端从哪里获得审批上下文？** 请求携带精确的 `callId` 和发起方的人类可读 `reason`；通道适配器可自行关联更丰富的工具调用状态，而无需在审批 seam 中重复携带参数。
|
||||
|
||||
|
||||
@@ -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: 42b78ad8341dd52c4dd146a2207a5ae909d28f1e
|
||||
2026-07-06-sandbox.zh.md: dfa3349e4d74d6f2c4944414c25fe3726d4a9b5a
|
||||
2026-07-06-sandbox.md: 4355ab57374f77f1733fff39e2fb4ccadcedaf6b
|
||||
2026-07-06-sandbox.zh.md: 02c5337555b7466c2bac7fc4774cdd2c945178ae
|
||||
|
||||
@@ -40,7 +40,7 @@ The swap is invisible to every consumer of `ctx.bash`: the bash tools, hook comm
|
||||
|
||||
Misconfiguration fails loud: `mode` outside the closed vocabulary is rejected at plugin load, and a host with no usable backend throws the structured `SANDBOX_UNAVAILABLE` — at `confine()` before the command ever spawns — rather than degrading to unconfined execution. `runnerCommand` on `dsh-sandbox-local` is the operator's explicit assertion of a bwrap-compatible runner (chain and probes skipped); it doubles as the deterministic fake-runner seam for keyless tests.
|
||||
|
||||
Denied file effects return a `[sandbox: file access denied under <mode> mode]` marker and instructions not to work around the denial. A confining executor adds paired `sandbox_permissions` and `justification` fields for one approved retry that must be strictly wider than the session's effective mode. A grant widens only that retry; rejection executes nothing, returns `the user rejected escalating this command to "<mode>"`, and permits no re-ask. The prompt does not announce sandbox mode, avoiding preemptive refusal. When `dsh-permission` is composed with a UI adapter, one preset selects both knob values; unmatched values fold to `custom`. The [ACP automation composition](../../../../examples/acp-agent/README.md) does not mount that UI service and selects its deployment mode explicitly.
|
||||
Denied file effects return a `[sandbox: file access denied under <mode> mode]` marker and instructions not to work around the denial. A confining executor adds paired `sandbox_permissions` and `justification` fields for one approved retry that must be strictly wider than the session's effective mode. A grant widens only that retry; rejection executes nothing, returns `the user rejected escalating this command to "<mode>"`, and permits no re-ask. The owner-derived runtime context states the current file policy without replacing those enforcement boundaries. When `dsh-permission` is composed with a UI adapter, one preset selects both knob values; unmatched values fold to `custom`. The [ACP automation composition](../../../../examples/acp-agent/README.md) does not mount that UI service and selects its deployment mode explicitly.
|
||||
|
||||
### Design detail
|
||||
|
||||
@@ -72,7 +72,7 @@ Backend profiles share the mode contract but differ in necessary host grants. La
|
||||
|
||||
`dsh-bash-sandbox` extends `LocalBashExecutor` and hands `ctx.sandbox` the exact `['bash', '-c', command]` argv it is about to spawn. A denial is an orthogonal result fact, conservatively classified from the active runner's stderr dialect. A runner failure outranks denial: foreground execution throws `SANDBOX_UNAVAILABLE`; a settled `BashProcess` stamps `sandbox.runnerFailed`, and the bash producer renders it through generic `task_output`.
|
||||
|
||||
The model's view is result facts only: the static tool description explains the denial marker (`[sandbox: file access denied under <mode> mode]`), encourages attempting commands that may be denied, and forbids retrying around a denial; when the escalation fields are advertised, a denied result additionally carries the escalation hint itself, so the sanctioned same-turn retry is prompted at the decision point rather than depending on the model recalling the description (§ Escalation). No prompt section states the sandbox mode (§ Per-session modes).
|
||||
The model sees the current effective file policy in the owner-derived `sandbox:policy` runtime context, while the static tool description explains the denial marker (`[sandbox: file access denied under <mode> mode]`), encourages attempting commands that may be denied, and forbids retrying around a denial; when the escalation fields are advertised, a denied result additionally carries the escalation hint itself, so the sanctioned same-turn retry is prompted at the decision point rather than depending on the model recalling the description (§ Escalation). [The current-policy decision](2026-07-30-current-sandbox-policy-context.md) owns the context's rationale and boundaries.
|
||||
|
||||
#### Escalation: one approved wider retry after a denial
|
||||
|
||||
@@ -105,7 +105,7 @@ interface SessionEventMap {
|
||||
|
||||
Each owner exports the same three-piece kit: the event declaration, a pure fold (`effectiveSandboxMode(events)` / `effectiveApprovalPolicy(events)` — a `findLast`, typed to the domain's closed union), and THE write path (`setSandboxMode(session, mode)` / `setApprovalPolicy(session, policy)` — a switch IS its event; nothing mutates state out of band). No shared owner service, no generic facts map, no registry: a third knob copies the ~40-line pattern into its own package. Execution follows the fold on both sides — the bash tool's per-call stamp reads it as the middle rung of the § Escalation precedence chain, and the approval seam's `'never'` gate is [the approval Agent Note](2026-07-06-approval-seam.md)'s side of the same pattern.
|
||||
|
||||
Sandbox mode is not narrated in the prompt; denial results report the mode when it matters, avoiding preemptive refusal based on a standing label. Approval policy is different: only `'never'` is stated because automatic rejection otherwise looks like a user decision. Policy-change notices are coalesced and delivered by the next pre-step, with log-derived fallback after restart. The notice source is inferred from event position: a knob event after the last request header is user-driven; unlogged drift is operator or config driven.
|
||||
Sandbox and approval policy are rendered as ordered contributions to one runtime-context snapshot before every request. The loop records the complete snapshot as a sourced `user/message`; both `'ask'` and `'never'` are explicit, so neither owner needs switch narration or last-told state.
|
||||
|
||||
**The optional UI surface** is `PermissionService`: a deployment-defined preset table whose entries bundle one sandbox mode with one approval policy. The shipped `workspace-write` and `danger-full-access` presets write through to both domain setters; a knob combination outside the table is reported as `custom`. UI adapters may expose that table as a selector. The automation-only ACP transport advertises no configuration selector and mounts no permission-preset service.
|
||||
|
||||
@@ -117,10 +117,10 @@ fs/web/todo execute in-process, so their sandbox semantics are policy at their s
|
||||
|
||||
### Testing
|
||||
|
||||
- **Unit:** pin platform selection and profiles, fail-closed runner classification, per-call mode/root resolution, per-process facts, escalation validation and outcomes, permission preset folding and write-through, and narrator coalescing.
|
||||
- **Unit:** pin platform selection and profiles, fail-closed runner classification, per-call mode/root resolution, per-process facts, escalation validation and outcomes, permission preset folding and write-through, and runtime-context ordering and materialization.
|
||||
- **Keyless real-runner:** exercise bwrap, Landlock, and Seatbelt against real filesystem effects at provider and bash-consumer layers; one real Cordis context concurrently drives two project sessions through shipped bash and fs tools, proving own-root success and sibling-root denial. Packed-install coverage proves the registry launcher remains executable. CI rejects a silent all-skip.
|
||||
- **With-key:** start the real ACP composition in read-only mode, let a model-driven bash write hit the runner's denial marker, then drive the bridge answerer and disk effect through granted and rejected workspace-write retries; unavailable credentials or runners self-skip.
|
||||
- **Snapshot:** pin prompt deltas and notices plus both scripted approval branches. A real ACP example scenario places its session under the user home while the deployment fallback points at `/tmp`, then pins a successful deployment-selected workspace-write mutation; this distinguishes session-root resolution from the process fallback without depending on runner-specific denial text. Other snapshots start unconfined so unrelated fixtures remain platform-independent.
|
||||
- **Snapshot:** pin the atomic current-policy context and both scripted approval branches. A real ACP example scenario places its session under the user home while the deployment fallback points at `/tmp`, then pins both the workspace-write runtime-context message and a successful deployment-selected mutation; this distinguishes session-root resolution from the process fallback without depending on runner-specific denial text. Other snapshots start unconfined so unrelated fixtures remain platform-independent.
|
||||
|
||||
## Deferred phases
|
||||
|
||||
@@ -149,9 +149,9 @@ Each phase gets its full design when picked up, validated against the code at th
|
||||
- **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.
|
||||
- **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 via `agent/user-message` + a bus event** — rejected: it presupposes a turn-entry seam that does not exist (the real seam is `agent/prompt-submit`), and pre-step's position serves both the coalesced turn-entry notice and the mid-turn immediacy bound with one listener.
|
||||
- **A standing prompt statement of the sandbox mode (+ a switch narrator)** — shipped first, then removed on live evidence: with `Bash commands run under the "read-only" file sandbox.` in every request, the model refused to ATTEMPT denied-then-escalatable work (five of twelve turns in the first manual session ended with zero tool calls), turning the sandbox into a soft lockout. The denial marker names the mode at the moment it matters and the escalation fields carry the recovery; the approval knob keeps its statement because an auto-rejection is behaviorally indistinguishable from a human "no".
|
||||
- **Track "last told" with its own bookkeeping events** — rejected: the `request/header` fold already records the exact prompt the model saw; parsing the closed candidate sentences back replaces a second bookkeeping stream — events are needed only where they ARE the store.
|
||||
- **Narrate each switch through `agent.inject()` plus a bus event** — rejected: independent notices expose owner ordering and intermediate combinations, while one assembly pass can materialize the complete current state atomically at the request boundary.
|
||||
- **State sandbox mode in the stable system prompt** — shipped first, then removed on live evidence: with `Bash commands run under the "read-only" file sandbox.` in every request, the model refused to ATTEMPT denied-then-escalatable work (five of twelve turns in the first manual session ended with zero tool calls), turning the sandbox into a soft lockout. The denial marker names the mode at the moment it matters and the escalation fields carry the recovery. The absence decision is superseded by [the current-policy decision](2026-07-30-current-sandbox-policy-context.md); this measurement and causal observation remain the evidence that any replacement must counter-test.
|
||||
- **Track "last told" with its own bookkeeping events** — rejected: the latest sourced runtime-context `user/message` records the exact full snapshot the model saw. Materializing that snapshot from current owner contributions replaces a second bookkeeping stream — events are needed only where they ARE the store.
|
||||
- **Independent sandbox and approval selectors** — rejected: one deployment-defined permission preset keeps the two policy knobs coherent for UI clients that expose runtime switching.
|
||||
|
||||
## Consequences
|
||||
@@ -160,12 +160,12 @@ What shipped pins — the tiers in Testing hold each:
|
||||
|
||||
- A denied command retried with `sandbox_permissions` + `justification` prompts the user through the composed answerer chain; a grant runs THAT call under the wider mode (result facts say so) while every other call keeps its own effective mode; every non-grant outcome produces its distinct error text and executes nothing.
|
||||
- The escalation fields exist exactly when the mounted executor confines; a request that is not strictly wider than the call's effective mode fails closed with its own text and prompts no one; a deployment with no ApprovalService fails escalating calls closed and leaves plain calls untouched.
|
||||
- The system prompt never states the sandbox mode (an approval `'never'` policy is the one stated knob), and the whole exchange — headers, knob events, notices, approvals, results — reconstructs from the session log alone, with no event types beyond the two knob events.
|
||||
- One preset selection records only changed knob values, while a no-op selection records nothing; an approval-policy switch is narrated in at most one coalesced notice, and a committed sandbox switch is honored by the next call's stamp.
|
||||
- A resumed session's overrides apply with no catch-up state; a default changed while the process was down is narrated before the session's first new request, attributed to the operator.
|
||||
- One sourced runtime-context message states the complete current sandbox and approval policies atomically; the whole exchange — context snapshots, headers, knob events, approval notices, approvals, and results — reconstructs from the session log alone, with no policy bookkeeping events beyond the two knob events.
|
||||
- One preset selection records only changed knob values, while a no-op selection records nothing; the next request snapshots both current values atomically, and a committed sandbox switch is honored by the next call's stamp.
|
||||
- A resumed session's overrides enter its first new runtime-context snapshot with no catch-up state; a composition default changed while the process was down likewise appears in that snapshot.
|
||||
- Two concurrent sessions never see each other's state or notices.
|
||||
- Two concurrent project sessions in one Cordis context resolve independent workspace roots; bash and fs writes succeed inside the calling session's cwd and fail against its neighbor's cwd.
|
||||
- `agent-loop` is untouched — everything rides `systemPrompt.section`, `SessionEventMap` merging, `agent.inject()`, `agent/step`, `agent/prompt-submit`, and capability-owned policy resolution.
|
||||
- Policy ownership stays in plugins through `systemPrompt.context`, `SessionEventMap` merging, and capability-owned resolution; the generic loop change materializes every owner's ordered context as one sourced message.
|
||||
|
||||
Costs and accepted limits:
|
||||
|
||||
@@ -178,9 +178,8 @@ Costs and accepted limits:
|
||||
- **The model may over-ask.** Escalating without denial grounding, or picking `danger-full-access` where `workspace-write` suffices: the description steers and the enum forces the ladder, but the human prompt is the actual gate; the `approval/asked` reasons make over-asking auditable, and a `prepend` policy answerer can auto-reject patterns a deployment never wants.
|
||||
- **The advertised target set is static while the effective mode is per-session** (schemas are registry-global) — a session already at the widest mode is still offered the fields. Harmless by construction: the strict-wider check at execution, not the enum, is the safety boundary — a non-widening request fails with its own text and never prompts anyone.
|
||||
- **A granted escalation is not a working sandbox.** An unavailable backend still fails closed even for a granted escalation to a confining mode — at `confine()` when the platform has no chain or every probe fails, at execution when an unprobed sole runner refuses (classified as a sandbox failure, not a command failure) — while a granted `danger-full-access` run never touches the provider at all: there the grant, not the probe, is the authority.
|
||||
- **The approval narrator's restart baseline parses prompt prose.** The closed candidate sentence is owned by the writing module itself, so a wording change is a coordinated writer+parser edit in one file; a session whose headers predate the section silently adopts the current policy without a notice.
|
||||
- **The approval section is still a dynamic prompt surface** (a `'never'` switch breaks provider prompt-prefix caching for that session). Accepted: policy switches are rare, and a model acting on a stale `'never'` is worse. The sandbox knob no longer touches the prompt at all.
|
||||
- **The model may hold a stale belief about the sandbox mode** (nothing announces a switch). Accepted deliberately: the next attempt's marker or success corrects it, and the observed failure mode of announcing — preemptive refusal — is worse than one wasted retry.
|
||||
- **Runtime-context history is append-only.** A policy switch appends a complete superseding snapshot after retained history, preserving the stable system-and-conversation prefix; unchanged state adds no message.
|
||||
- **Older policy snapshots remain in history.** Each full snapshot explicitly supersedes earlier runtime-context snapshots, so replay and compaction need only retain the latest materialized message.
|
||||
|
||||
## FAQ
|
||||
|
||||
@@ -191,8 +190,8 @@ Costs and accepted limits:
|
||||
- **Does the sandbox restrict network or process visibility?** No — `SandboxMode` claims FILE effects only; the bwrap profile deliberately does not unshare pid, and no backend claims network. Whether network restriction becomes its own knob is left open in § The seam.
|
||||
- **Which tools actually run confined?** OS subprocesses through `ctx.bash` — the bash tools, and hook commands transitively — plus the filesystem tools (`read`/`write`/`edit`) through the sandboxed `ctx.fs` provider (the [cross-family fs sandbox RFC](2026-07-14-cross-family-fs-sandbox.md)): bash confines via the OS runner, fs via an in-process path fence, both keying off the same `ctx.sandboxPolicy` mode. web/todo stay in-process and unfenced (web's only effect is network, outside the file-effect mode vocabulary).
|
||||
- **Does a granted escalation persist?** No. The grant is consumed by the exact foreground or background call that asked; every neighboring call keeps its own effective mode. A later background denial surfaces through `task_output` and may ground a new exact-command retry.
|
||||
- **When does a runtime mode switch take effect?** Once its session event commits, the very next capability resolution folds and stamps the new mode. The model is not told a standing mode; its next command simply behaves under the new policy, and any denial names that policy at the point of use.
|
||||
- **What survives a restart — and what if the operator changed the config default while the process was down?** Overrides replay from the session log (`effective = fold ?? config`), so a resumed session keeps its modes with zero catch-up machinery; a default that drifted offline changes behavior the same way a switch does (the approval policy, being stated, is additionally narrated with operator/config attribution).
|
||||
- **When does a runtime mode switch take effect?** Once its session event commits, the next runtime-context snapshot and the next capability resolution fold the new mode. The sourced context message records what the model was told, and any later denial names the same policy at the point of use.
|
||||
- **What survives a restart — and what if the operator changed the config default while the process was down?** Overrides replay from the session log (`effective = fold ?? config`), so a resumed session keeps its modes with zero catch-up machinery; a default that drifted offline enters the next full runtime-context snapshot.
|
||||
- **What does `enforcement: 'partial'` on a result mean?** The selected backend enforces the subset its kernel ABI governs — e.g. Landlock before ABI v3 does not govern path truncate — and says so structurally instead of refusing the host; the probe's report line distinguishes the cases. The bwrap and Seatbelt profiles govern every promised file effect by construction, so they always report `full`.
|
||||
|
||||
## Prior art
|
||||
|
||||
@@ -40,7 +40,7 @@ harness 是一个 SDK，因此约束必须是开发者可组合的能力：是
|
||||
|
||||
配置错误大声失败：`mode` 不在封闭词汇中时在插件加载时被拒绝；主机上没有可用后端时在 `confine()` 阶段（命令 spawn 之前）抛出结构化的 `SANDBOX_UNAVAILABLE`，而非降级为无约束执行。`dsh-sandbox-local` 上的 `runnerCommand` 是运维人员对一个 bwrap 兼容 runner 的显式断言（跳过链和探测）；它同时充当 keyless 测试的确定性 fake-runner seam。
|
||||
|
||||
被拒绝的文件操作返回 `[sandbox: file access denied under <mode> mode]` 标记，并附带不要绕过拒绝的指令。约束执行器添加配对的 `sandbox_permissions` 和 `justification` 字段，用于一次经批准的重试，该重试必须严格宽于会话的有效模式。授权仅放宽该次重试；拒绝则不执行任何内容，返回 `the user rejected escalating this command to "<mode>"`，且不允许再次请求。提示词不声明沙箱模式，以避免基于常驻标签的预防性拒绝。当 `dsh-permission` 与某个 UI 适配器一起组合时，一个 preset 同时选定两个旋钮值；不匹配的组合折叠为 `custom`。[ACP 自动化组合](../../../../examples/acp-agent/README.md)不挂载该 UI 服务，而是显式选定其部署模式。
|
||||
被拒绝的文件操作返回 `[sandbox: file access denied under <mode> mode]` 标记，并附带不要绕过拒绝的指令。约束执行器添加配对的 `sandbox_permissions` 和 `justification` 字段，用于一次经批准的重试，该重试必须严格宽于会话的有效模式。授权仅放宽该次重试；拒绝则不执行任何内容，返回 `the user rejected escalating this command to "<mode>"`，且不允许再次请求。由归属方派生的运行时上下文会说明当前文件策略，但不会取代这些强制执行边界。当 `dsh-permission` 与某个 UI 适配器一起组合时，一个 preset 同时选定两个旋钮值；不匹配的组合折叠为 `custom`。[ACP 自动化组合](../../../../examples/acp-agent/README.md)不挂载该 UI 服务，而是显式选定其部署模式。
|
||||
|
||||
### 设计细节
|
||||
|
||||
@@ -72,7 +72,7 @@ Landlock launcher 源码和包工作区位于 `native/landlock-run`，与 harnes
|
||||
|
||||
`dsh-bash-sandbox` 扩展 `LocalBashExecutor`，并把即将 spawn 的确切 `['bash', '-c', command]` argv 交给 `ctx.sandbox`。拒绝是与其他结果正交的事实，依据当前 runner 的 stderr 方言保守分类。Runner 失败优先于拒绝：前台执行抛出 `SANDBOX_UNAVAILABLE`；结算后的 `BashProcess` 会盖章 `sandbox.runnerFailed`，bash 生产者再通过通用 `task_output` 渲染它。
|
||||
|
||||
模型看到的仅是结果事实：静态工具描述解释拒绝标记（`[sandbox: file access denied under <mode> mode]`），鼓励尝试可能被拒绝的命令，并禁止绕过拒绝重试；当升级字段被公布时，被拒绝的结果还额外携带升级提示本身，使被认可的同轮次重试在决策点被提示，而非依赖模型回忆描述（§ 升级机制）。没有提示词段落声明沙箱模式（§ 按会话模式）。
|
||||
模型会在归属方派生的 `sandbox:policy` 运行时上下文中看到当前有效的文件策略；静态工具描述则解释拒绝标记（`[sandbox: file access denied under <mode> mode]`），鼓励尝试可能被拒绝的命令，并禁止绕过拒绝重试。当升级字段被公布时，被拒绝的结果还会携带升级提示本身，使被认可的同轮次重试在决策点获得提示，而非依赖模型回忆描述（§ 升级机制）。[当前策略决策](2026-07-30-current-sandbox-policy-context.md)负责该上下文的理由与边界。
|
||||
|
||||
#### 升级机制：拒绝后一次经批准的更宽重试
|
||||
|
||||
@@ -105,7 +105,7 @@ interface SessionEventMap {
|
||||
|
||||
每个拥有者导出相同的三件套：事件声明、纯 fold（`effectiveSandboxMode(events)` / `effectiveApprovalPolicy(events)`——一个 `findLast`，类型化到领域的封闭联合），以及唯一的写入路径（`setSandboxMode(session, mode)` / `setApprovalPolicy(session, policy)`——切换即其事件；没有任何东西在带外修改状态）。无共享拥有者服务、无通用 facts map、无注册表：第三个旋钮只需将约 40 行模式复制到自己的包中。执行在两侧都遵循 fold——bash 工具的按调用盖章将其作为 § 升级机制优先级链的中间层读取，approval seam 的 `'never'` 门控是[批准 Agent Note](2026-07-06-approval-seam.md) 同一模式的另一侧。
|
||||
|
||||
沙箱模式不在提示词中叙述；拒绝结果在需要时报告模式，避免基于常驻标签的预防性拒绝。批准策略不同：只有 `'never'` 被声明，因为自动拒绝在行为上与用户的「不」无法区分。策略变更通知被合并，由下一个步骤前检查点递送，重启后有基于日志的回退。通知来源从事件位置推断：最后一个 request header 之后的旋钮事件是用户驱动的；未记录的漂移是运维人员或配置驱动的。
|
||||
沙箱策略与批准策略会在每次请求前渲染为同一份运行时上下文快照中的有序贡献。循环会将完整快照记录为一条带来源的 `user/message`；`'ask'` 与 `'never'` 都会明确写入，因此两个归属方都无需切换叙述或「上次告知」状态。
|
||||
|
||||
**可选的 UI 界面**是 `PermissionService`：一张部署定义的 preset 表，每个条目捆绑一个沙箱模式与一个批准策略。随附的 `workspace-write` 和 `danger-full-access` preset 写入两个领域 setter；preset 表之外的旋钮组合报告为 `custom`。UI 适配器可以把该表暴露为选择器。仅面向自动化的 ACP 传输层不公布任何配置选择器，也不挂载权限 preset 服务。
|
||||
|
||||
@@ -117,10 +117,10 @@ fs/web/todo 在进程内执行，因此它们的沙箱语义是各自 seam 层
|
||||
|
||||
### 测试
|
||||
|
||||
- **单元测试：** 固定平台选择和 profile、失败关闭的 runner 分类、按调用的模式/根目录解析、按进程事实、升级验证和结果、权限 preset fold 和写入透传、以及叙述器合并。
|
||||
- **单元测试：** 固定平台选择和 profile、失败关闭的 runner 分类、按调用的模式/根目录解析、按进程事实、升级验证和结果、权限 preset fold 和写入透传、以及运行时上下文排序与具体化。
|
||||
- **Keyless 真实 runner：** 在提供方和 bash 消费方层面对 bwrap、Landlock 和 Seatbelt 执行真实文件系统效果测试；一个真实 Cordis 上下文通过已交付的 bash 和 fs 工具并发驱动两个项目会话，证明在自身根目录写入成功、在兄弟根目录写入被拒绝。Packed-install 覆盖率证明注册表 launcher 保持可执行。CI 拒绝静默全跳过。
|
||||
- **With-key：** 以只读模式启动真实 ACP 组合，让模型驱动的 bash 写入命中 runner 的拒绝标记，再通过已授权与被拒绝的 workspace-write 重试驱动 bridge 应答器和磁盘效果；不可用的凭证或 runner 自动跳过。
|
||||
- **快照：** 固定提示词 delta 和通知，以及两个脚本化的 approval 分支。一个真实 ACP 示例场景把会话放在用户主目录下，同时让部署后备根目录指向 `/tmp`，然后固定一次成功的、由部署选定的 workspace-write 变更；这能区分会话根目录解析与进程后备值，而不依赖 runner 特定的拒绝文本。其他快照以无约束启动，使无关 fixture（测试前置数据）保持平台无关。
|
||||
- **快照：** 固定原子化的当前策略上下文和两个脚本化的 approval 分支。一个真实 ACP 示例场景把会话放在用户主目录下，同时让部署后备根目录指向 `/tmp`，然后固定 workspace-write 运行时上下文消息与一次成功的、由部署选定的变更；这能区分会话根目录解析与进程后备值，而不依赖 runner 特定的拒绝文本。其他快照以无约束启动，使无关 fixture（测试前置数据）保持平台无关。
|
||||
|
||||
## 延迟阶段
|
||||
|
||||
@@ -149,9 +149,9 @@ fs/web/todo 在进程内执行，因此它们的沙箱语义是各自 seam 层
|
||||
- **按会话动态工具 schema**：否决。schema 设计上是注册表全局的（一套 assembly 词汇、固定 header 快照契约），按会话重新注册只能买到执行时严格放宽检查已保证的东西，代价是按会话的 schema 表面和每次切换的 header 变动。
|
||||
- **将重试硬匹配到先前的拒绝**：否决。命令字符串同一性脆弱（引号、`workdir`、env 前缀、作为失败阶段重试的管道）——要么误拒诚实的重试，要么被轻易满足；真正的边界是人看到命令 + 理由。仅在 `allow_always` 授权存储需要机器可检查的范围时才重新考虑。
|
||||
- **通用 `env/state` facts map 加拥有者服务**：否决。approval 和沙箱独立组合，因此任何一方的状态都不应拖入第三个包；单键 fold 各自是一个 `findLast`，拥有者服务自然消解；没有跨旋钮的不变式，因此原子多键补丁无收益。
|
||||
- **通过 `agent/user-message` + 总线事件叙述**：否决。它预设了一个不存在的轮次入口 seam（真正的 seam 是 `agent/prompt-submit`），而步骤前检查点的位置使一个监听器能够同时服务合并的轮次入口通知和轮中即时性约束。
|
||||
- **提示词中常驻声明沙箱模式（+ 切换叙述器）**：先交付后移除，基于实际证据：当每个请求中都有 `Bash commands run under the "read-only" file sandbox.` 时，模型拒绝尝试被拒绝后可升级的工作（首次手动会话中十二个轮次有五个以零工具调用结束），将沙箱变成了软锁定。拒绝标记在需要时命名模式，升级字段承载恢复路径；批准旋钮保留其声明，因为自动拒绝在行为上与人的「不」无法区分。
|
||||
- **用专门的簿记事件追踪「上次告知」**：否决。`request/header` fold 已记录模型看到的确切提示词；将封闭的候选句子解析回来替代了第二条簿记流——事件仅在它们本身即为存储时才需要。
|
||||
- **通过 `agent.inject()` 加总线事件逐次叙述切换**：否决。独立通知会暴露归属方顺序和中间组合，而一次组装过程可以在请求边界以原子方式具体化完整的当前状态。
|
||||
- **在稳定系统提示词中声明沙箱模式**：先行交付，随后根据线上证据移除：每次请求都带有 `Bash commands run under the "read-only" file sandbox.` 时，模型会拒绝尝试本可在被拒后升级的工作（首次人工会话的十二个轮次中有五个以零工具调用结束），使沙箱变成软锁死。拒绝标记会在相关时刻指出模式，升级字段则承载恢复路径。[当前策略决策](2026-07-30-current-sandbox-policy-context.md)取代了省略策略的决策；这项测量和因果观察仍是任何替代方案必须进行反证测试的依据。
|
||||
- **用专门的簿记事件追踪「上次告知」**：否决。最新一条带来源的运行时上下文 `user/message` 会记录模型看到的确切完整快照。根据当前归属方贡献具体化该快照，取代了第二条簿记流——事件仅在它们本身即为存储时才需要。
|
||||
- **相互独立的沙箱与批准选择器**：否决。一个部署定义的权限 preset 让两个策略旋钮对暴露运行时切换的 UI 客户端保持一致。
|
||||
|
||||
## 后果
|
||||
@@ -160,12 +160,12 @@ fs/web/todo 在进程内执行，因此它们的沙箱语义是各自 seam 层
|
||||
|
||||
- 被拒绝的命令以 `sandbox_permissions` + `justification` 重试时，通过组合的应答器链提示用户；授权使该次调用在更宽模式下运行（结果事实如此报告），而其他所有调用保持各自的有效模式；每种非授权结果产生各自不同的错误文本且不执行任何内容。
|
||||
- 升级字段恰好在已挂载的执行器约束时存在；不严格宽于调用有效模式的请求以自身文本失败关闭且不提示任何人；没有 ApprovalService 的部署对升级调用失败关闭，对普通调用不影响。
|
||||
- 系统提示词从不声明沙箱模式（批准 `'never'` 策略是唯一被声明的旋钮），且整个交互——header、旋钮事件、通知、批准、结果——仅从会话日志即可重建，除两个旋钮事件外无额外事件类型。
|
||||
- 一次 preset 选择只记录发生变化的旋钮值，而无操作的选择不记录任何内容；批准策略切换最多以一条合并通知叙述，已提交的沙箱切换由下一次调用的盖章兑现。
|
||||
- 恢复的会话的覆盖直接生效，无需追赶状态；进程停止期间变更的默认值在会话的首个新请求前被叙述，归因于运维人员。
|
||||
- 一条带来源的运行时上下文消息会以原子方式声明完整的当前沙箱策略与批准策略；整个交互——上下文快照、header、旋钮事件、批准通知、批准与结果——仅从会话日志即可重建，除两个旋钮事件外没有策略簿记事件。
|
||||
- 一次 preset 选择只记录发生变化的旋钮值，而无操作的选择不记录任何内容；下一个请求会把两个当前值共同纳入一份原子快照，已提交的沙箱切换由下一次调用的盖章兑现。
|
||||
- 恢复会话的覆盖项会进入其首个新运行时上下文快照，无需追赶状态；进程停止期间变更的组合默认值也会出现在该快照中。
|
||||
- 两个并发会话永远看不到彼此的状态或通知。
|
||||
- 同一个 Cordis 上下文中的两个并发项目会话解析各自独立的工作区根目录；bash 和 fs 写入在调用方会话的 cwd 内成功，对其相邻会话的 cwd 则失败。
|
||||
- `agent-loop` 未被触及——一切搭载 `systemPrompt.section`、`SessionEventMap` 合并、`agent.inject()`、`agent/step`、`agent/prompt-submit` 和由能力拥有的策略解析。
|
||||
- 策略归属仍通过 `systemPrompt.context`、`SessionEventMap` 合并和由能力归属方拥有的解析留在插件中；通用循环变更会将每个归属方的有序上下文具体化为一条带来源的消息。
|
||||
|
||||
代价与已接受的限制：
|
||||
|
||||
@@ -178,9 +178,8 @@ fs/web/todo 在进程内执行，因此它们的沙箱语义是各自 seam 层
|
||||
- **模型可能过度请求。** 在没有拒绝依据的情况下升级，或在 `workspace-write` 足够时选择 `danger-full-access`：描述引导且枚举强制阶梯，但人的提示词是实际门控；`approval/asked` 原因使过度请求可审计，且 `prepend` 策略应答器可以自动拒绝部署永远不想要的模式。
|
||||
- **公布的目标集是静态的，而有效模式是按会话的**（schema 是注册表全局的）——已处于最宽模式的会话仍被提供这些字段。构造上无害：执行时的严格放宽检查（而非枚举）是安全边界——非放宽请求以自身文本失败且不提示任何人。
|
||||
- **授权的升级不等于可工作的沙箱。** 不可用的后端即使对授权升级到约束模式也仍然失败关闭——在平台没有链或所有探测失败时于 `confine()` 阶段，在未探测的唯一 runner 拒绝时于执行阶段（归类为沙箱失败而非命令失败）——而授权的 `danger-full-access` 运行根本不触及提供方：此时授权（而非探测）是权威。
|
||||
- **批准叙述器的重启基线解析提示词文本。** 封闭的候选句子由写入模块本身拥有，因此措辞变更是同一文件中写入器+解析器的协调编辑；header 早于该段落的会话静默采用当前策略而不发通知。
|
||||
- **批准段落仍是动态提示词表面**（`'never'` 切换会破坏该会话的提供方提示词前缀缓存）。已接受：策略切换罕见，且模型基于过时的 `'never'` 行动更糟。沙箱旋钮不再触及提示词。
|
||||
- **模型可能持有关于沙箱模式的过时信念**（没有任何东西宣布切换）。有意接受：下一次尝试的标记或成功会纠正它，而宣布的观察到的失败模式——预防性拒绝——比一次浪费的重试更糟。
|
||||
- **运行时上下文历史仅追加。** 策略切换会在保留的历史之后追加一份用于取代先前快照的完整快照，从而保留稳定的系统与对话前缀；状态不变时不添加消息。
|
||||
- **旧策略快照仍保留在历史中。** 每份完整快照都会明确取代更早的运行时上下文快照，因此回放与压缩（compaction）只需保留最新具体化的消息。
|
||||
|
||||
## FAQ
|
||||
|
||||
@@ -191,8 +190,8 @@ fs/web/todo 在进程内执行，因此它们的沙箱语义是各自 seam 层
|
||||
- **沙箱限制网络或进程可见性吗？** 不——`SandboxMode` 仅声称文件操作；bwrap profile 刻意不 unshare pid，没有后端声称网络。网络限制是否成为自己的旋钮留在 § seam 中开放。
|
||||
- **哪些工具实际在约束下运行？** 通过 `ctx.bash` 的 OS 子进程——bash 工具及传递性的钩子命令——再加上通过沙箱化 `ctx.fs` 提供方运行的文件系统工具（`read`/`write`/`edit`，见[跨工具族 fs 沙箱 Agent Note](2026-07-14-cross-family-fs-sandbox.md)）：bash 通过 OS runner 约束，fs 通过进程内路径围栏约束，二者都以同一个 `ctx.sandboxPolicy` 模式为键。web/todo 仍在进程内且不受限制（web 的唯一效果是网络，不在文件效果模式词汇内）。
|
||||
- **授权的升级会持久化吗？** 不会。授权由发起请求的确切前台或后台调用消费；每个相邻调用保留自己的有效模式。后续的后台拒绝通过 `task_output` 呈现，并且可以作为一次新的精确命令重试的依据。
|
||||
- **运行时模式切换何时生效？** 一旦其会话事件提交，紧接着的下一次能力解析就会折叠并盖章新模式。模型不被告知常驻模式；其下一个命令直接在新策略下运行，任何拒绝都会在使用点命名该策略。
|
||||
- **重启后什么存活——如果运维人员在进程停止期间改了配置默认值呢？** 覆盖从会话日志回放（`effective = fold ?? config`），因此恢复的会话以零追赶机制保持其模式；离线漂移的默认值以与切换相同的方式改变行为（批准策略因被声明，还额外以运维人员/配置归因叙述）。
|
||||
- **运行时模式切换何时生效？** 一旦其会话事件提交，下一个运行时上下文快照与下一次能力解析都会折叠新模式。带来源的上下文消息会记录模型收到的内容，之后的任何拒绝都会在使用点命名同一策略。
|
||||
- **重启后什么存活——如果运维人员在进程停止期间改了配置默认值呢？** 覆盖从会话日志回放（`effective = fold ?? config`），因此恢复的会话以零追赶机制保持其模式；离线漂移的默认值会进入下一份完整运行时上下文快照。
|
||||
- **结果上的 `enforcement: 'partial'` 是什么意思？** 所选后端强制其内核 ABI 管控的子集——例如 ABI v3 之前的 Landlock 不管控路径 truncate——并以结构化方式如此声明而非拒绝主机；探测的报告行区分各种情况。bwrap 和 Seatbelt profile 构造上管控所有承诺的文件操作，因此始终报告 `full`。
|
||||
|
||||
## 先例
|
||||
|
||||
@@ -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/2026-07-30-current-sandbox-policy-context.md
|
||||
2026-07-30-current-sandbox-policy-context.md: 093f9a8a0b58256bb4f2301cc62fa83629de4fbd
|
||||
2026-07-30-current-sandbox-policy-context.zh.md: 729ea9d50c7dbbfa027e1cad11929ae1c2b379b1
|
||||
@@ -0,0 +1,63 @@
|
||||
# Agent Note: Current sandbox policy context
|
||||
|
||||
Status: implemented
|
||||
|
||||
English | [中文](2026-07-30-current-sandbox-policy-context.zh.md)
|
||||
|
||||
## Problem
|
||||
|
||||
The sandbox policy already enforced and logged each session's file-effect mode, but a fresh model request did not contain that state. In a Web session under `read-only`, write and edit schemas remained visible, so the model claimed it could write and learned otherwise only after a denied call. After `/permission danger-full-access`, the next request carried the approval-policy change but still omitted the sandbox mode. Denial results were therefore the first model-visible policy source even when the user asked about capability before any operation.
|
||||
|
||||
## Decision
|
||||
|
||||
`dsh-sandbox-policy`, the owner of mode and workspace-root resolution, registers one `sandbox:policy` cache-safe context contribution. Every agent request resolves the active session directly through `ctx.sandboxPolicy.resolve({ session })`; there is no denial-history scan or process-local “last told” state.
|
||||
|
||||
The policy contribution is capability-neutral and present for every agent session. It does not maintain a second inventory of mounted backends or tools; model-visible schemas remain the authority for available operations, while the context conditions its claims on any available operation that the DSH file sandbox enforces. The [capability-neutral policy context decision](../simplification/2026-07-31-capability-neutral-sandbox-policy-context.md) supersedes the earlier family-registration mechanism while retaining this note's cache-safe delivery and durable snapshot design.
|
||||
|
||||
The contribution states only facts shared by every enforcement dialect. `read-only` says an available sandbox-enforced operation cannot modify files in the standing mode and directs the model to try an available tool normally, then follow any denial and escalation guidance that tool returns. `workspace-write` states the canonical session workspace with non-exclusive wording and summarizes, without enumerating, that some platform temporary areas may also be writable. `danger-full-access` says the DSH file sandbox does not restrict file modifications by available operations. Backend-selected temporary paths, `/dev/null`, runner readiness, exact tool availability, and other policy domains are absent because `resolve()` cannot establish them at request assembly.
|
||||
|
||||
The existing `dsh-system-prompt` assembly now has ordered dynamic contexts alongside stable system sections and tool schemas. After assembling one step, agent-loop renders all active contexts as one full snapshot with an explicit supersession statement. It appends a sourced `user/message` only when no retained snapshot exists, the bytes changed, compaction removed the retained message, or the final contribution disappeared and needs one clearing snapshot. The snapshot is appended after existing history and before `step/start`, so a changed policy preserves the preceding system-and-conversation cache prefix. The session event itself reconstructs the exact model input; `request/header` remains byte-identical when only policy context changes.
|
||||
|
||||
Ownership stays narrow. Approval policy contributes its complete current `ask` or `never` fact to the same full snapshot; migrating sandbox alone would not preserve cache because `/permission` changes both owners. Plan mode remains `plan:policy`, and tool plugins continue to own schemas plus attempt, denial, and escalation guidance. Context states standing policy; filesystem, one-shot bash, and terminal backends remain the enforcement boundaries.
|
||||
|
||||
The cache decision follows current source rather than analogy alone. Codex models permissions as a developer-role `WorldState` section with a persisted fingerprint, emits it only when state changes or retained history lost the fragment, and records the snapshot transition. Hermes keeps its system prompt fixed for a session and explicitly prepends changing skill, model, and voice notices to the next user message to avoid invalidating prompt cache. Pi has no comparable built-in sandbox state, and Claude Code's current native implementation is not publicly inspectable; Anthropic's public cache guidance nevertheless places changing per-request context after the stable cached prefix.
|
||||
|
||||
The earlier real-provider Web fixture quantified the defect in the system-section version. The first `danger-full-access` and `workspace-write` requests each reported only 256 cache-read tokens against 14,691 and 14,782 uncached input tokens. Later steps under an unchanged policy reported approximately 14.7k–15.5k cache-read tokens. Moving only the sandbox sentence would not fix those misses because the same preset switch also rewrote the approval-policy system section.
|
||||
|
||||
## Wording evidence
|
||||
|
||||
The wording experiment pre-registered preemptive refusal as its primary endpoint and required the old standing sentence to produce at least one refusal in twelve fresh sessions before any replacement could be judged. On 2026-07-30, commit `2bf41990401b194bd8637f07bbd90c67a9eeac75` ran `deepseek-v4-flash` through the shipped Web composition with the exact positive-control sentence `Bash commands run under the "read-only" file sandbox.` and the current tool-owned attempt guidance. The control produced zero preemptive refusals and zero speculative escalations; all twelve sessions made an ordinary bash call, observed a denial, escalated in the same turn, received approval, and landed the requested file. No sample was excluded.
|
||||
|
||||
After the cache-safe delivery change, commit `10d4e0ff7b68d38fc4403403b644aac442b97a00` repeated the same twelve-session positive control through the new tail-context channel. It again produced zero preemptive refusals and zero speculative escalations; all twelve sessions made an ordinary first call, observed denial, escalated in the same turn, and received approval. Eight landed the exact requested file, and no sample was excluded.
|
||||
|
||||
Both positive controls therefore failed the pre-registered sensitivity gate. The formal twelve-session Candidate A and B arms were not run, and these experiments do not select or validate the current wording. They establish that the earlier five-of-twelve result is not reproducible under this task and current tool guidance, and that a stronger positive control or different task distribution is required before making model-behavior rate claims. Deterministic tests below establish truthful request construction and replay only.
|
||||
|
||||
The cache-safe delivery rework then supplied a separate, non-statistical acceptance comparison over the neutral Web task `Create the relative path policy-neutral.txt ...`; it does not replace the pre-registered twelve-session experiment. Candidate A's categorical read-only statement produced a text refusal with zero tool calls. Candidate B added one composition-conditioned sentence only for enforced families whose tools expose escalation. A fresh real-provider run then issued an ordinary `write`, observed the read-only denial, retried the same operation in the same turn with `sandbox_permissions: "workspace-write"`, received approval, read the file back, and verified the exact contents. It made no speculative escalation. Across the permission switches and four mutation steps, cache reads were 14,848–15,872 tokens while uncached input was 59–306 tokens per request, directly demonstrating the stable-prefix benefit.
|
||||
|
||||
## Alternatives considered
|
||||
|
||||
**Narrate only mode changes.** Rejected because it leaves a fresh session uninformed and makes the first denied operation the policy-discovery mechanism. It also requires a baseline definition that is unnecessary when current state can be rendered directly.
|
||||
|
||||
**Scan denial history or remember the last narrated mode.** Rejected because denial events describe attempted operations, not authoritative current state, while process-local bookkeeping does not survive resume. The owner can fold the durable policy directly on every request.
|
||||
|
||||
**Put current policy in a dynamic system section.** Rejected after real provider evidence showed that a first-time permission switch reduced cache reads to 256 tokens while roughly 14.7k input tokens missed. DeepSeek matches complete prefixes; changing the first wire message prevents reuse of the longer system-plus-history prefix.
|
||||
|
||||
**Call `agent.inject()` independently from each policy owner.** Rejected because sibling listener order would define model order, separate messages could expose mismatched intermediate snapshots, and every owner would need its own compaction-retention scan. The existing assembly owner can order contributions and materialize one atomic full snapshot.
|
||||
|
||||
**A generic runtime-facts package.** Rejected because the existing system-prompt assembly already owns sections, schemas, variables, scope, and the authoritative per-step waterfall. Extending that owner with ordered contexts adds no package or second registry service.
|
||||
|
||||
**Repeat tool schemas or plan guidance in the context.** Rejected because those surfaces already have owners and independent lifecycles. Approval current state joins the snapshot only because the same `/permission` switch changes it and leaving its system section would retain the cache defect.
|
||||
|
||||
**Keep Candidate A after the cache-safe move.** Rejected by the neutral real-provider task: the model returned a pure text refusal and made no tool call despite the existing bash attempt guidance. The surviving anti-refusal principle states no escalation mechanics itself; it tells the model not to infer impossibility from the standing label, then delegates denial and escalation behavior back to the available tool.
|
||||
|
||||
**Keep sandbox mode absent because a standing mode label once caused preemptive refusal.** Rejected because a fresh Web request otherwise exposes mutation tools while withholding their standing policy, producing false capability claims before the first operation. The earlier live measurement remains a required counter-test: five of twelve turns ended without a tool call under `Bash commands run under the "read-only" file sandbox.` The committed tool-owned attempt guidance postdates that measurement, so the replacement is selected through a new positive-control experiment under the current tool contract rather than assuming the old and current conditions match.
|
||||
|
||||
**A separate model-context package.** Rejected because the policy owner can resolve current session state directly and the existing assembly service can order it. A new package would add a shallow composition seam and documentation/gate surface around the same request boundary.
|
||||
|
||||
**Enumerate writable temporary roots.** Rejected because the backend is selected later at `confine()`: bwrap, Landlock, Seatbelt, and the in-process filesystem fence do not grant one common temporary-path set. Host-specific paths in a standing request would be both unstable and overclaimed.
|
||||
|
||||
## Consequences
|
||||
|
||||
A model receives the standing file policy before probing a tool, and the next request after `/permission` reflects the committed mode. The stable system prompt no longer changes for sandbox or approval state; a changed full context snapshot is append-only after retained history, and unchanged state adds no message. Older snapshots remain in history but are explicitly superseded by the latest full snapshot. The statement is guidance, not an enforcement guard: runtime safety still comes from filesystem, one-shot bash, and terminal backends consuming the same resolved policy.
|
||||
|
||||
Focused tests pin all modes, canonical roots, switch timing, service disposal, context ordering, clearing, stable request headers, resume, and byte stability across different `TMPDIR` values. Keyless assembled snapshots pin the durable context message through real Loader compositions. Keyless replay owns the neutral denial-to-escalation trajectory; it is a structural regression proof, not wording-selection evidence.
|
||||
@@ -0,0 +1,63 @@
|
||||
# Agent Note: 当前沙箱策略上下文
|
||||
|
||||
Status: implemented
|
||||
|
||||
[English](2026-07-30-current-sandbox-policy-context.md) | 中文
|
||||
|
||||
## 问题
|
||||
|
||||
沙箱策略已经强制执行并记录每个会话的文件操作模式，但新的模型请求并不包含这一状态。在 `read-only` 下的 Web 会话中，write 与 edit schema 仍然可见，因此模型会声称自己能够写入，直到一次被拒绝的调用后才发现事实并非如此。执行 `/permission danger-full-access` 后，下一个请求带有批准策略变更，却仍省略沙箱模式。因此，即使用户在任何操作前询问能力，拒绝结果也是模型可见的首个策略来源。
|
||||
|
||||
## 决策
|
||||
|
||||
`dsh-sandbox-policy` 负责解析模式与工作区根目录，并注册一项缓存安全的 `sandbox:policy` 上下文贡献。每次 agent（智能体）请求都通过 `ctx.sandboxPolicy.resolve({ session })` 直接解析当前会话；不存在拒绝历史扫描或进程本地的「上次告知」状态。
|
||||
|
||||
该策略贡献不依赖具体能力，并存在于每个 agent 会话中。它不会另行维护一份已挂载后端或工具清单；模型可见的 schema 仍是可用操作的权威来源，而上下文会将其声明限定在 DSH 文件沙箱所强制执行的任何可用操作上。[不依赖具体能力的策略上下文决策](../simplification/2026-07-31-capability-neutral-sandbox-policy-context.md)取代了较早的家族注册机制，同时保留本 Agent Note 的缓存安全交付与持久快照设计。
|
||||
|
||||
该贡献只说明所有强制执行方言所共有的事实。`read-only` 表明受沙箱强制执行的可用操作在常驻模式下无法修改文件，并指示模型正常尝试可用工具，随后遵循该工具返回的任何拒绝与升权引导。`workspace-write` 用非排他措辞说明规范化的会话工作区，并概述某些平台临时区域可能也可写，而不逐一列举。`danger-full-access` 表明 DSH 文件沙箱不会限制可用操作修改文件。后端选择的临时路径、`/dev/null`、runner 就绪状态、确切的工具可用情况和其他策略领域都不会出现，因为 `resolve()` 无法在请求组装时确定它们。
|
||||
|
||||
现有 `dsh-system-prompt` 组装在稳定系统段与工具 schema 之外，还包含有序的动态上下文。组装一个步骤后，agent loop（智能体循环）会将所有活动上下文渲染成一份带显式取代声明的完整快照。仅当不存在保留快照、字节发生变化、压缩（compaction）移除了保留消息，或最后一项贡献消失而需要一份清除快照时，它才会追加一条带来源的 `user/message`。快照追加在现有历史之后、`step/start` 之前，因此策略变化时仍会保留此前的系统与对话缓存前缀。会话事件本身可以重建确切的模型输入；只有策略上下文变化时，`request/header` 仍逐字节相同。
|
||||
|
||||
归属范围保持收敛。批准策略会将其完整的当前 `ask` 或 `never` 事实贡献给同一份完整快照；只迁移沙箱无法保留缓存，因为 `/permission` 会同时改变两方。计划模式仍由 `plan:policy` 负责，工具插件也继续负责各自的 schema，以及尝试、拒绝与升级引导。上下文负责说明常驻策略；文件系统、一次性 bash 与终端后端仍是强制执行边界。
|
||||
|
||||
缓存决策依据当前源码，而不只依靠类比。Codex 将权限建模为 developer 角色的 `WorldState` 段并保存其指纹；只有状态变化或保留的历史丢失该片段时才发出它，同时记录快照转换。Hermes 在会话期间保持系统提示词不变，并明确将不断变化的 skill（技能）、模型与语音通知前置到下一条用户消息，以免提示词缓存失效。Pi 没有可比的内置沙箱状态，Claude Code 当前的原生实现也无法公开检视；不过，Anthropic 的公开缓存指南仍将不断变化的逐请求上下文放在稳定缓存前缀之后。
|
||||
|
||||
先前接入真实提供方的 Web fixture（测试前置数据）量化了系统段版本的缺陷。首次 `danger-full-access` 和 `workspace-write` 请求分别只有 256 个缓存读取 token，而未缓存输入 token 为 14,691 和 14,782 个。相同策略下的后续步骤报告约 14.7k–15.5k 个缓存读取 token。只移动沙箱语句无法修复这些未命中，因为同一次 preset 切换还会改写批准策略系统段。
|
||||
|
||||
## 措辞证据
|
||||
|
||||
措辞实验预先登记「预防性拒绝」为主要终点，并要求旧常驻句子在十二个 fresh session 中至少产生一次拒绝，之后才能评判任何替代措辞。2026-07-30，commit `2bf41990401b194bd8637f07bbd90c67a9eeac75` 通过已交付的 Web 组合运行 `deepseek-v4-flash`，使用精确的阳性对照句子 `Bash commands run under the "read-only" file sandbox.` 与当前工具归属方的尝试引导。对照组产生零次预防性拒绝和零次推测性升级；十二个会话全部先发起普通 bash 调用、观察到拒绝、在同一轮次升级、获得批准，并让所请求文件实际落盘。没有样本被排除。
|
||||
|
||||
缓存安全交付变更后，commit `10d4e0ff7b68d38fc4403403b644aac442b97a00` 通过新的尾部上下文通道重复了同一项十二会话阳性对照。结果再次为零次预防性拒绝和零次推测性升权；十二个会话的首次调用均为普通调用，随后观察到拒绝、在同一轮次升权并获得批准。其中八个会话让所请求文件按确切要求落盘，没有样本被排除。
|
||||
|
||||
因此，两项阳性对照均未通过预先登记的灵敏度门槛。Candidate A 与 B 的正式十二会话实验组均未运行，这些实验不选择也不验证当前措辞。它们说明先前十二次中五次的结果无法在本任务与当前工具引导下复现；在声明模型行为率之前，需要更强的阳性对照或不同的任务分布。下述确定性测试只证明请求构造与回放真实一致。
|
||||
|
||||
随后，缓存安全交付重做针对中性 Web 任务 `Create the relative path policy-neutral.txt ...` 提供了一次独立的非统计验收对比；它不取代预先登记的十二会话实验。Candidate A 的绝对化只读声明导致模型以纯文本拒绝，工具调用为零。Candidate B 只针对受强制执行、且其工具公开升权能力的家族增加一句按组合条件化的文案。随后一次全新的真实提供方运行先发出普通 `write`，观察到只读拒绝，再在同一轮次用 `sandbox_permissions: "workspace-write"` 重试同一操作，获得批准、读回文件并核验确切内容。它没有进行推测性升权。在权限切换和四个变更步骤中，每个请求的缓存读取为 14,848–15,872 个 token，未缓存输入为 59–306 个 token，直接证明了稳定前缀的收益。
|
||||
|
||||
## 曾考虑的替代方案
|
||||
|
||||
**仅叙述模式变更。** 不予采用，因为这会让新会话不了解策略，并把首次被拒绝的操作变成策略发现机制。如果可以直接渲染当前状态，也就无需额外定义基线。
|
||||
|
||||
**扫描拒绝历史或记住上次叙述的模式。** 不予采用，因为拒绝事件描述的是尝试过的操作，而不是权威的当前状态；进程本地的簿记也无法跨恢复保留。归属方可以在每次请求时直接折叠持久策略。
|
||||
|
||||
**把当前策略放入动态系统段。** 不予采用，因为真实提供方证据显示，首次权限切换后缓存读取降至 256 个 token，而约 14.7k 个输入 token 未命中缓存。DeepSeek 匹配完整前缀；改变第一条 wire 消息会阻止复用更长的系统与历史前缀。
|
||||
|
||||
**由每个策略归属方独立调用 `agent.inject()`。** 不予采用，因为同级监听器的顺序会决定模型所见顺序，分开的消息可能暴露不匹配的中间快照，并且每个归属方都需要各自扫描压缩后的保留状态。现有组装归属方可以对贡献排序，并具体化一份原子化的完整快照。
|
||||
|
||||
**通用运行时事实包。** 不予采用，因为现有系统提示词组装已经拥有段、schema、变量、作用域和权威的逐步骤 waterfall（瀑布式事件）。为该归属方增加有序上下文，无需新增包或第二个注册表服务。
|
||||
|
||||
**在上下文中重复工具 schema 或计划引导。** 不予采用，因为这些接口已有各自归属方和独立生命周期。批准的当前状态加入快照，仅仅是因为同一个 `/permission` 切换会改变它，而把它留在系统段会保留缓存缺陷。
|
||||
|
||||
**缓存安全迁移后仍保留 Candidate A。** 不予采用，因为中性的真实提供方任务中，尽管已有 bash 尝试引导，模型仍以纯文本拒绝，且没有调用工具。保留下来的反预防性拒绝原则本身不说明任何升权机制；它告诉模型不要从常驻标签推断操作不可能完成，然后把拒绝与升权行为交还给可用工具。
|
||||
|
||||
**继续省略沙箱模式，因为常驻模式标签曾引发预防性拒绝。** 不予采用，因为新的 Web 请求否则会暴露变更工具，却隐去这些工具的常驻策略，导致模型在首次操作前错误声称自身能力。先前的线上测量仍是必须执行的反证测试：使用 `Bash commands run under the "read-only" file sandbox.` 时，十二个轮次中有五个没有调用工具。已提交的工具归属方尝试引导晚于该测量，因此应通过当前工具契约下的新阳性对照实验选择替代文案，而不能假设旧条件与当前条件相同。
|
||||
|
||||
**独立的模型上下文包。** 不予采用，因为策略归属方可以直接解析当前会话状态，现有组装服务也可以对其排序。新包只会围绕同一个请求边界引入浅层组合 seam 和额外的文档／门禁表面。
|
||||
|
||||
**枚举可写临时根目录。** 不予采用，因为后端要到稍后的 `confine()` 才会选定：bwrap、Landlock、Seatbelt 和进程内文件系统围栏并不授予一套共同的临时路径。常驻请求中的主机特定路径既不稳定，也会作出过度承诺。
|
||||
|
||||
## 后果
|
||||
|
||||
模型在试探工具前就会收到常驻文件策略，且 `/permission` 后的下一个请求会反映已提交的模式。稳定的系统提示词不再随沙箱或批准状态变化；变化后的完整上下文快照会在保留的历史之后仅追加，状态不变时不增加消息。较旧的快照仍保留在历史中，但最新的完整快照会明确取代它们。该声明是引导，而不是强制执行护栏：运行时安全仍来自文件系统、一次性 bash 与终端后端消费同一项解析完成的策略。
|
||||
|
||||
聚焦测试固定了所有模式、规范化根目录、切换时机、服务释放、上下文顺序、清除、稳定的请求 header、恢复，以及不同 `TMPDIR` 值下的字节稳定性。无密钥的组装快照通过真实 Loader 组合固定持久上下文消息。无密钥回放负责固定中性的拒绝到升级轨迹；它是结构回归证明，而不是措辞选型证据。
|
||||
@@ -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/simplification/2026-07-31-capability-neutral-sandbox-policy-context.md
|
||||
2026-07-31-capability-neutral-sandbox-policy-context.md: d45c6c652fb7a061c2f92b7e03c41cce39da2a30
|
||||
2026-07-31-capability-neutral-sandbox-policy-context.zh.md: 9a1bb7a13d934826dddf89b409b25ba892a31ea3
|
||||
@@ -0,0 +1,35 @@
|
||||
# Agent Note: Capability-neutral sandbox policy context
|
||||
|
||||
Status: implemented
|
||||
|
||||
English | [中文](2026-07-31-capability-neutral-sandbox-policy-context.zh.md)
|
||||
|
||||
## Problem
|
||||
|
||||
The current-policy context originally mirrored runtime composition through separate enforced-family and escalatable-family registries. Six backend, tool, and example call sites contributed `filesystem`, `bash`, or `terminal`; the policy service retained token sets for independent disposal, intersected and ordered the two registries, invalidated prompt assemblies on every lifecycle change, and tested every family combination.
|
||||
|
||||
That inventory was neither needed to state the file policy nor authoritative for model-visible capability. A backend contribution could name a family whose model-facing tool was absent or hidden by request scope, while tool schemas already told the model which exact operations were available. The registries therefore widened the public service and lifecycle contract to maintain an approximate English sentence.
|
||||
|
||||
## Decision
|
||||
|
||||
`dsh-sandbox-policy` contributes one capability-neutral `sandbox:policy` context for every agent session. It derives the text only from `resolve({ session })`; there is no backend or tool family registration API, contribution map, ordering rule, or registration-driven prompt invalidation.
|
||||
|
||||
The text conditions capability claims on available operations that the DSH file sandbox enforces. Under `read-only`, it states that such operations cannot modify files in the standing mode and tells the model to try an available tool normally, then follow any denial and escalation guidance that tool returns. Under `workspace-write`, it states the canonical session workspace and the qualified temporary-area allowance. Under `danger-full-access`, it states that the DSH file sandbox does not restrict file modifications by available operations.
|
||||
|
||||
Tool schemas remain the authority for which operations are available. Tool results remain the authority for operation-specific denials and approved wider retries. Filesystem, one-shot bash, and terminal implementations continue to resolve and enforce the same per-call policy; only the redundant model-facing capability inventory is removed.
|
||||
|
||||
This decision partially supersedes the family-registration and composition-conditioned wording in [the current sandbox policy context decision](../feature/2026-07-30-current-sandbox-policy-context.md). That note remains the owner of cache-safe context delivery, durable snapshot materialization, wording evidence, and the separation between guidance and enforcement.
|
||||
|
||||
## Alternatives considered
|
||||
|
||||
**Keep the registries but reduce their tests.** Rejected because the public methods, retained lifecycle state, six contribution sites, and approximate capability claim would remain. The combinatorial tests reflected the design cost; they did not create it.
|
||||
|
||||
**Derive an exact inventory from the tool registry.** Rejected because current policy needs only a truthful conditional statement, while exact availability already appears in the assembled tool schemas and can vary by request scope. Mapping each schema back to an enforcement backend would introduce another derived relation with no current consumer.
|
||||
|
||||
**Omit policy context unless a backend advertises enforcement.** Rejected because it recreates the registration problem and makes policy visibility depend on optional contributors. The conditional wording remains truthful even when no applicable operation is available.
|
||||
|
||||
## Consequences
|
||||
|
||||
The policy service has one owner-derived context path instead of two public registries and their disposal lifecycle. Capability additions and removals no longer churn the runtime-context snapshot, while mode and workspace changes still do. Exact mode wording, canonical roots, switching, resume, service disposal, and no-agent assembly remain covered; family-combination and contribution-lifecycle tests disappear with the behavior they protected.
|
||||
|
||||
The model no longer receives a prose list of sandboxed capability families. It receives exact tool schemas plus one standing file-policy statement. If a future product needs a separate capability inventory, it must be derived from the authoritative per-request assembly rather than reconstructed through backend registration side channels.
|
||||
@@ -0,0 +1,35 @@
|
||||
# Agent Note: 不依赖具体能力的沙箱策略上下文
|
||||
|
||||
Status: implemented
|
||||
|
||||
[English](2026-07-31-capability-neutral-sandbox-policy-context.md) | 中文
|
||||
|
||||
## 问题
|
||||
|
||||
当前策略上下文最初通过受强制执行家族与可升权家族两个独立注册表来映射运行时组合。后端、工具与示例中的六个调用点会贡献 `filesystem`、`bash` 或 `terminal`；策略服务保留 token 集合，以便独立释放各项注册，对两个注册表的内容求交集并排序，在每次生命周期变化时使提示词组装失效，并且需要测试覆盖所有家族组合。
|
||||
|
||||
这份清单既不是说明文件策略所必需的，也不能作为模型可见能力的权威依据。后端贡献可能声明某个家族，而该家族面向模型的工具并不存在，或被请求作用域隐藏；工具 schema 已经向模型准确说明哪些操作可用。因此，为了维护一句近似的英文说明，注册表扩大了公开服务与生命周期契约。
|
||||
|
||||
## 决策
|
||||
|
||||
`dsh-sandbox-policy` 为每个 agent（智能体）会话贡献一项不依赖具体能力的 `sandbox:policy` 上下文。上下文文本只根据 `resolve({ session })` 派生；不存在后端或工具家族注册 API、贡献映射、排序规则，也不会因注册变化而使提示词失效。
|
||||
|
||||
该文本将能力声明限定于 DSH 文件沙箱所强制执行的可用操作。在 `read-only` 下，它说明这类操作在常驻模式下无法修改文件，并指示模型正常尝试可用工具，随后遵循该工具返回的任何拒绝与升权引导。在 `workspace-write` 下，它说明规范化的会话工作区，以及带有适用条件的临时区域写入许可。在 `danger-full-access` 下，它说明 DSH 文件沙箱不会限制可用操作修改文件。
|
||||
|
||||
哪些操作可用仍以工具 schema 为准。针对具体操作的拒绝以及获批后的更宽松模式重试，仍以工具结果为准。文件系统、一次性 bash 与终端实现继续解析并强制执行相同的逐调用策略；移除的只有面向模型的冗余能力清单。
|
||||
|
||||
本决策取代了[当前沙箱策略上下文决策](../feature/2026-07-30-current-sandbox-policy-context.md)中关于家族注册与按组合条件化措辞的部分内容。缓存安全的上下文交付、持久快照具体化、措辞证据，以及引导与强制执行之间的边界，仍以该 Agent Note 为归属文档。
|
||||
|
||||
## 曾考虑的替代方案
|
||||
|
||||
**保留注册表，但减少对应测试。** 不予采用，因为公开方法、保留的生命周期状态、六个贡献点和近似的能力声明仍会存在。组合测试只是体现了这一设计成本，并非成本的成因。
|
||||
|
||||
**从工具注册表派生确切清单。** 不予采用，因为当前策略只需要一句符合事实的条件性声明，而确切的可用情况已经体现在组装后的工具 schema 中，并且可能随请求作用域变化。若再把每个 schema 映射回其强制执行后端，就会引入另一项派生关系，而当前没有消费方需要它。
|
||||
|
||||
**仅在后端声明会强制执行时提供策略上下文。** 不予采用，因为这会重现注册问题，并使策略是否可见取决于可选贡献方。即使没有可适用的操作，这段带条件的措辞仍然符合事实。
|
||||
|
||||
## 后果
|
||||
|
||||
策略服务只有一条由归属方直接派生的上下文路径，无需维护两个公开注册表及其释放生命周期。增加或移除能力不再导致运行时上下文快照反复变化，模式和工作区变化仍会触发快照更新。聚焦测试仍覆盖各模式的确切措辞、规范化根目录、模式切换、恢复、服务释放与无 agent 组装；家族组合与贡献生命周期测试则随其所保护的行为一同删除。
|
||||
|
||||
模型不再收到以自然语言列出的沙箱能力家族清单，而是收到确切的工具 schema 与一项常驻文件策略声明。未来若产品需要独立的能力清单，该清单必须从权威的逐请求组装结果中派生，不能通过后端注册这种旁路机制重建。
|
||||
Reference in New Issue
Block a user