Coder/deepseek-harness
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

Merge remote-tracking branch 'origin/worktree-agent-scope-design' into codex/pr224-simplification-audit

Browse Source 
# Conflicts:
#	docs/config-catalog.md
#	docs/cordis-catalog/services.md
This commit is contained in:
Tianyi Cui
2026-07-12 23:57:01 +08:00
parent 84a9b7374c f7b9cea733
commit f0ea495b97
169 changed files with 8917 additions and 7170 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
packages/ui/acp/README.md
View File

@@ -36,7 +36,7 @@ The `initialize` handshake reports a fixed server identity (`agentInfo: { name:
## Multi-session
The bridge multiplexes N sessions over one connection. Live sessions are held in a `Map<sessionId, SessionRecord>` (forward) with a `WeakMap<Agent, sessionId>` reverse map so `agent/*` events — which carry only the `Agent` — demux in O(1). Every `session/event` and `agent/status` is routed strictly to its owning record, so concurrent sessions never cross-settle or interleave their `session/update` notifications. State is per session: one in-flight prompt each, `session/cancel` aborts and settles only its own agent/prompt, and disposal drains every live session in parallel to quiescence. Permission prompts follow the same ownership: the `approval/request` answerer resolves the owning session through the reverse map and prompts only there.
The bridge multiplexes N sessions over one connection. Live sessions are held in a `Map<sessionId, SessionRecord>` (forward) with a `WeakMap<Agent, sessionId>` reverse map so agent-scoped approval events demultiplex in O(1). Every `session/event` is routed strictly to its owning record, so concurrent sessions never cross-settle or interleave their `session/update` notifications. State is per session: one in-flight prompt each, `session/cancel` aborts and settles only its own agent/prompt, and disposal drains every live session in parallel to quiescence. Permission prompts follow the same ownership: the `approval/request` answerer resolves the owning session through the reverse map and prompts only there.
## Session config options
@@ -71,7 +71,7 @@ When the client does NOT advertise the capability, none of the `_meta`/terminal
## Settle-exactly-once
A `session/prompt` resolves (or rejects) exactly once, keyed off the canonical session log (the `session/event` stream). One listener captures the prompt's owning turn from the log's `turn/start` and settles on the matching `turn/end` — the durable boundary event (`closeTurn` appends it unconditionally; there is no `agent/*` turn mirror). A prompt settles only on ITS OWN turn (`inflight.turn === turn/end.turn`), so a stale `turn/end` for a previously-cancelled turn whose end arrives late can never settle the wrong prompt. A turn that ends `error` REJECTS the RPC with an internal error carrying the failure message (ACP has no error stop reason); every other reason resolves via the codec. As a fallback, when the agent settles to `idle`/`disposed` with a prompt still pending — e.g. a peer `session/event` listener registered before the bridge threw and starved the bridge's listener — an `agent/status` handler reconciles the prompt from the log (the owning turn's `turn/end`, or `cancelled` if the turn was torn down without one). An empty/whitespace prompt is rejected up front — it would queue no work, so no turn would start and the RPC would hang.
A `session/prompt` resolves or rejects exactly once from the canonical `session/event` stream. The listener captures the prompt's owning turn from `turn/start` and settles in a `finally` block when the matching `turn/end` is appended, so a presentation/streaming failure cannot strand the RPC after the durable terminal event exists. Correlation by turn id prevents a late end from a cancelled prompt from settling its successor. A turn ending in `error` rejects the RPC with an internal error carrying the failure message because ACP has no error stop reason; every other reason resolves through the codec. An empty or whitespace-only prompt is rejected before enqueue because it would start no turn and otherwise leave the RPC pending.
## Permission prompts

132
packages/ui/acp/src/index.ts
View File

@@ -71,7 +71,7 @@ import {
} from '@agentclientprotocol/sdk'
import type { ContentBlock } from '@deepseek-ai/dsh-llm'
import { assertNever, CallId } from '@deepseek-ai/dsh-llm'
import type { Agent, AgentStatus } from '@deepseek-ai/dsh-agent'
import type { Agent } from '@deepseek-ai/dsh-agent'
import { AgentId } from '@deepseek-ai/dsh-agent'
import { SessionId } from '@deepseek-ai/dsh-session'
import { SANDBOX_MODES, effectiveSandboxMode, setSandboxMode } from '@deepseek-ai/dsh-bash'
@@ -297,7 +297,8 @@ interface SessionRecord {
/**
* The in-flight `session/prompt`, or `undefined` when none is pending. A
* prompt resolves with a {@link StopReason} or rejects with an Error (a
* turn that ended in failure). Settled exactly once via {@link settlePrompt}.
* turn that ended in failure). Settled exactly once by its matching
* `turn/end`, direct cancellation, or teardown.
*
* `turn` is the loop turn number this prompt owns, captured from the log's
* `turn/start` after `send()`. Until then it is `undefined` (the turn has not
@@ -307,18 +308,11 @@ interface SessionRecord {
* wrong prompt. A direct cancel/dispose settle clears the whole in-flight slot,
* so a later stale `turn/end` finds no pending prompt.
*
* `logWatermark` is the session log length at the moment the prompt was
* installed (before `send()`). The settle-from-log fallback uses it to infer
* the owning `turn/start` from the canonical log even when the live
* `session/event` capture was starved (a peer listener that throws on
* `turn/start` — see `settleFromLog`): the prompt owns the FIRST `turn/start`
* appended at or after this watermark.
*/
inflight: {
resolve: (reason: StopReason) => void
reject: (error: Error) => void
turn: number | undefined
logWatermark: number
} | undefined
/**
* Config switches accepted while the session was IDLE, not yet anchored in
@@ -337,13 +331,9 @@ interface SessionRecord {
/**
* Drive the in-flight prompt's settle from the harness event stream. The bridge
* settles off the durable log: the `turn/end` session event on the
* `session/event` feed for the prompt's own turn, with the agent
* erroring/settling to idle as a fallback (docs/defensive-patterns.md "honor
* cross-seam contracts on BOTH sides") for the case where a throwing peer `session/event` listener
* starved the bridge's listener before it saw the boundary. The first of these
* to fire settles the prompt; `settle` is then cleared so the others are no-ops
* (settle-exactly-once).
* settles off the durable `turn/end` event for the prompt's own turn. Session
* contains post-commit observers independently, and this listener performs
* correlation in a `finally` so presentation failure cannot starve settlement.
*/
export function apply(ctx: Context, config: AcpConfig): void {
// Capture the injected services NOW, during apply(), while we are inside this
@@ -503,83 +493,24 @@ export function apply(ctx: Context, config: AcpConfig): void {
ctx.on('session/event', (session, event: SessionEvent) => {
const rec = sessions.get(session.header.id)
if (rec === undefined) return
streamSessionEventUpdate(rec.sessionId, event, notify, rec.presenter, {
enabled: rec.terminalEnabled,
cwd: session.header.cwd,
}, { includeUserMessages: false })
const inflight = rec.inflight
if (inflight === undefined) return
if (event.type === 'turn/start') {
// Tag the in-flight prompt with its owning turn — but ONLY a
// `message`-triggered turn (the kind a `send()` prompt produces). A turn
// a plugin opens between prompt-install and the prompt's own turn (an idle
// `agent.inject()` writes a one-shot `injection`-triggered turn) must NOT
// be mistaken for the prompt's turn, or its turn/end would settle the RPC
// early. The first message turn at/after install owns the prompt
// (`turn === undefined` guard); the loop batches queued messages into one
// turn, so there is exactly one.
if (inflight.turn === undefined && event.data.trigger.kind === 'message') {
inflight.turn = event.data.turn
try {
streamSessionEventUpdate(rec.sessionId, event, notify, rec.presenter, {
enabled: rec.terminalEnabled,
cwd: session.header.cwd,
}, { includeUserMessages: false })
} finally {
const inflight = rec.inflight
if (inflight !== undefined && event.type === 'turn/start') {
// The first message-triggered turn after prompt installation owns the
// prompt; injection-triggered turns must not settle it early.
if (inflight.turn === undefined && event.data.trigger.kind === 'message') {
inflight.turn = event.data.turn
}
} else if (inflight !== undefined && event.type === 'turn/end' && inflight.turn === event.data.turn) {
rec.inflight = undefined
settleFromTurnEnd(inflight, event.data.reason)
}
return
}
// Settle only on the OWNING turn's end.
if (event.type !== 'turn/end' || inflight.turn !== event.data.turn) return
rec.inflight = undefined
settleFromTurnEnd(inflight, event.data.reason)
})
// Settle fallback: a `session/event` listener registered BEFORE ACP that
// throws (on `turn/start` OR `turn/end`) would, via cordis `emit`'s
// stop-on-throw, starve ACP's listener above — the prompt would hang or, if
// only the turn number was missed, settle as the wrong outcome. So when the
// agent settles to `idle` (or is disposed), reconcile against the canonical
// log: determine the prompt's owning turn (the captured `turn`, or — if the
// live capture was starved — the FIRST `turn/start` appended at/after the
// install-time `logWatermark`), then settle from that turn's `turn/end`
// (reject on error, resolve via codec), or `cancelled` if no owning turn ever
// started. Never double-settles — clears `inflight` first.
const settleFromLog = (rec: SessionRecord): void => {
const inflight = rec.inflight
if (inflight === undefined) return
const events = rec.agent.session.events
// The owning turn number: the captured one, or — if the live capture was
// starved — inferred from the log as the first MESSAGE-triggered turn opened
// at/after the watermark. The message-trigger filter matches the live
// capture: a one-shot `injection` turn a plugin may open between
// prompt-install and the prompt's turn is NOT the prompt's turn. Undefined
// only if no message turn ever started for this prompt.
const owningTurn = inflight.turn ?? events.slice(inflight.logWatermark).find(
(e): e is Extract<SessionEvent, { type: 'turn/start' }> =>
e.type === 'turn/start' && e.data.trigger.kind === 'message',
)?.data.turn
// The owning turn's end in the log. If `owningTurn` is undefined (no turn
// ever started for this prompt — a torn-down-before-turn case that quiesce's
// direct settle normally pre-empts), no `turn/end` matches (turn numbers are
// >= 1) and `findLast` returns undefined, falling through to cancelled.
const end = events.findLast(
(e): e is Extract<SessionEvent, { type: 'turn/end' }> =>
e.type === 'turn/end' && e.data.turn === owningTurn,
)
rec.inflight = undefined
if (end === undefined) {
// No owning turn / no clean turn/end (torn down mid-turn) → cancelled.
inflight.resolve('cancelled')
return
}
settleFromTurnEnd(inflight, end.data.reason)
}
// On a settle to idle/disposed, reconcile any still-pending prompt from the
// log (covers a starved `session/event` listener — see settleFromLog). A mid-
// step disposal that never appended a clean turn/end resolves `cancelled`.
// Demux via the agent→sessionId reverse map.
ctx.on('agent/status', (agent, status: AgentStatus) => {
const sessionId = bySession.get(agent)
if (sessionId === undefined) return
const rec = sessions.get(sessionId)
if (rec === undefined) return
if (status === 'idle' || status === 'disposed') settleFromLog(rec)
})
// --- Approval answerer -----------------------------------------------------
@@ -904,12 +835,10 @@ export function apply(ctx: Context, config: AcpConfig): void {
// Install the in-flight slot BEFORE send() (send does not synchronously
// flip status to running; the session/event listener records the turn
// number and settle/rejects it). Capture the log length now as the
// watermark: the settle-from-log fallback infers the owning turn/start
// as the first one appended at/after it, surviving a starved live
// capture. A turn that ends in error rejects this promise (the codec
// never produces an error stop reason).
// A turn that ends in error rejects this promise (the codec never
// produces an error stop reason).
const stopReason = await new Promise<StopReason>((resolve, reject) => {
rec.inflight = { resolve, reject, turn: undefined, logWatermark: rec.agent.session.events.length }
rec.inflight = { resolve, reject, turn: undefined }
rec.agent.send([{ type: 'text', text }])
})
return { stopReason }
@@ -928,8 +857,7 @@ export function apply(ctx: Context, config: AcpConfig): void {
// as cancelled directly here: do NOT rely on the resulting turn/end to
// settle it, because cancel() may drop the turn before any turn/end is
// emitted, and removing this direct settle would move the RPC's
// resolution onto the settleFromLog/agent-status path, changing its
// timing.
// resolution onto a later observer path, changing its timing.
rec.agent.cancel('session/cancel')
settlePrompt(rec, 'cancelled')
return Promise.resolve()
@@ -1012,15 +940,15 @@ export function apply(ctx: Context, config: AcpConfig): void {
* quiescence"): for each session settle any pending prompt `cancelled`, then
* run that session's {@link AgentHandle} `dispose()` — which stops the loop
* (sets `disposed`, aborts the in-flight step), AWAITS the loop's exit (the
* final `turn/end` + `session/flush` are captured while `onAppend` is still
* final `turn/end` + `session/flush` are captured while the store-owned publication hooks are still
* attached), unregisters the agent, and removes its session from the store.
* The per-session disposes run in parallel. Idempotent — clears the `sessions`
* map first and memoizes, so a second call (close racing dispose) is a no-op.
* Shared by Cordis disposal AND client disconnect (`conn.closed`).
*
* Per-agent disposal closes the former pre-step best-effort window — but via
* the DISPOSED path, not `cancel()`: the start-disposer resolves `handle.disposed`,
* which wakes the parked loop, and `isDisposed()` breaks the loop before a
* Per-agent disposal closes the queued-before-run window through the DISPOSED
* path, not `cancel()`: the start-disposer resolves `handle.disposed`, which
* wakes the parked loop, and `isDisposed()` breaks the loop before a
* queued-but-not-yet-running turn can start (a turn cut off mid-flight ends
* with reason `disposed`, not `aborted`). A bare client disconnect (resolves
* `conn.closed` WITHOUT disposing the fiber) thus leaves NO registered agent

8
packages/ui/acp/tests/dispose.spec.ts
View File

@@ -159,8 +159,8 @@ describe('acp bridge — disposal & HMR safety', () => {
it('the final turn closing events are persisted across an AgentHandle dispose (durability)', async () => {
// The teardown-ORDER guarantee: a per-agent dispose must stop the loop,
// AWAIT its exit (so the loop's final `turn/end` + `session/flush` fire
// through the still-attached `session.onAppend` → `session/event`), and only
// THEN detach onAppend + remove the session. If the order were inverted
// through the still-attached store observer → `session/event`), and only
// THEN remove its publication hooks and session entry. If the order were inverted
// (detach first), the closing events would never reach persistence. Drive a
// CLEAN turn to completion, dispose JUST the bridge, then re-load the
// persisted log from disk and assert the closing turn/end is on disk — the
@@ -190,7 +190,7 @@ describe('acp bridge — disposal & HMR safety', () => {
// produced BY the dispose itself. Here the model stream HANGS, so the turn is
// still open when teardown runs: the composite agent effect stops the loop,
// the loop unwinds and appends `turn/end {disposed}` + runs its final
// `session/flush` — all while `onAppend` is still attached (the session
// `session/flush` — all while the store-owned publication hooks are still attached (the session
// detach is the LAST disposer in the same effect's LIFO chain) — and only
// THEN is the session detached. If the order were inverted (or the session
// were a racing SIBLING effect), the abort-produced `turn/end` would never
@@ -255,7 +255,7 @@ describe('acp bridge — disposal & HMR safety', () => {
// into ONE composite effect whose disposers run as a `.then()` chain. The
// register disposer emits `agent/disposed`; if a listener throws and the
// emit is UNCONTAINED, the rejected chain skips the LATER session-detach
// disposer — stranding the session in the store with `onAppend` attached (a
// disposer — stranding the session in the store with its publication hooks attached (a
// leak AND a durability hole, since the new design relies on detach
// running). The emit must be contained. Register a throwing listener, drive
// a clean turn, dispose, and assert the session was STILL removed.

2
packages/ui/acp/tests/edges.spec.ts
View File

@@ -20,7 +20,7 @@ describe('acp bridge — demux & config edges', () => {
it('ignores events from an agent the bridge does not own (strict id demux)', async () => {
// A second agent created directly on the registry (NOT via the bridge) runs
// a turn. Its session/event + agent/status must NOT produce ACP updates and
// a turn. Its session events must NOT produce ACP updates and
// must not settle anything — the bridge demuxes strictly by its own id.
harness = await makeBridgeHarness({ storageDir, script: [textResponse('foreign')] })
await harness.client.initialize({ protocolVersion: PROTOCOL_VERSION, clientCapabilities: {} })

30
packages/ui/acp/tests/turns.spec.ts
View File

@@ -235,12 +235,9 @@ describe('acp bridge — turn outcomes', () => {
expect(failed).toHaveLength(1)
})
it('settles via the log fallback when a prior session/event listener throws (starvation)', async () => {
// A peer session/event listener that runs BEFORE the bridge's listener
// throws on turn/end (prepend: true puts it first). cordis emit stops at the
// throw, so the bridge's session/event listener never sees turn/end and
// cannot settle there. The agent/status idle-fallback must reconcile the
// prompt from the log so the RPC settles instead of hanging.
it('settles successfully when an earlier turn/end observer throws', async () => {
// Session contains each post-commit observer failure, so a prepended peer
// cannot starve the bridge's live turn/end delivery.
harness = await makeBridgeHarness({ storageDir, script: [textResponse('answer')] })
harness.ctx.on('session/event', (_s, event) => {
if (event.type === 'turn/end') throw new Error('peer listener boom')
@@ -250,9 +247,7 @@ describe('acp bridge — turn outcomes', () => {
expect(res.stopReason).toBe('end_turn')
})
it('log fallback REJECTS when the starved turn ended in error', async () => {
// Same starvation as above, but the turn fails: the idle-fallback must
// reject the RPC from the logged turn/end{error}, not resolve.
it('still rejects a failed turn when an earlier turn/end observer throws', async () => {
harness = await makeBridgeHarness({ storageDir, script: [errorResponse('starved boom')] })
harness.ctx.on('session/event', (_s, event) => {
if (event.type === 'turn/end') throw new Error('peer listener boom')
@@ -262,21 +257,18 @@ describe('acp bridge — turn outcomes', () => {
.rejects.toThrow(/turn failed: starved boom/)
})
it('log fallback infers the owning turn when turn/START capture is starved', async () => {
// A peer listener throws on turn/START (not turn/end): the bridge never
it('captures and settles the owning turn when an earlier turn-start observer throws', async () => {
// Turn correlation still reaches the bridge after the throwing peer and
// captures inflight.turn via the live stream. A throwing turn/start listener
// also FAILS the turn (the throw is recorded as the turn's error). Without
// the watermark inference the fallback would resolve `cancelled` (the bug);
// with it, it infers the owning turn from the log and REJECTS from that
// turn's error turn/end. (The model's own error is never reached — the turn
// failed at start — so the rejection carries the listener's failure.)
harness = await makeBridgeHarness({ storageDir, script: [textResponse('never runs')] })
// Session contains post-commit callbacks independently.
// The model request and normal turn outcome therefore still occur.
harness = await makeBridgeHarness({ storageDir, script: [textResponse('answer')] })
harness.ctx.on('session/event', (_s, event) => {
if (event.type === 'turn/start') throw new Error('peer listener boom on start')
}, { prepend: true })
const sessionId = await newSession(harness)
await expect(harness.client.prompt({ sessionId, prompt: [{ type: 'text', text: 'go' }] }))
.rejects.toThrow(/turn failed:/)
const result = await harness.client.prompt({ sessionId, prompt: [{ type: 'text', text: 'go' }] })
expect(result.stopReason).toBe('end_turn')
})
it('a between-turn injection does not settle the prompt early (message-trigger correlation)', async () => {

4
packages/ui/user-approval/README.md
View File

@@ -2,11 +2,11 @@
User-approval seam. Owns the `ctx.approval` service ([`ApprovalService`](src/index.ts)) and the one-shot permission vocabulary the harness shares: `ApprovalRequest` (agent + tool identity + reason + abort signal), the closed `ApprovalOutcome` union (`allowed-once` / `rejected` / `cancelled` / `unavailable`), the `ApprovalRequestId` brand pairing the two log-only audit events (`approval/asked` / `approval/decided`), and the `approval/request` waterfall the answerers listen on. It lives in the UI group because its purpose is human permission, while remaining channel-neutral: it depends only on Cordis and core vocabulary packages, never on a concrete UI.
The contract in one line: `ctx.approval.request(req)` puts exactly one question — "may this specific action proceed?" — to whatever answerers the deployment composed, and always resolves to an outcome, never rejects: an aborted signal yields `cancelled`, a throwing or missing answerer yields `unavailable`, and `allowed-once` is a grant for the single asked-about action, never a class of future ones. Acceptance is synchronous: the service shallow-freezes a detached request record before dispatch, preserving the exact `agent` and `AbortSignal` identities while making later caller mutation unable to redirect scope, payload, cancellation, or either audit event. Session observers run after an event enters the append-only log; if one throws, the service recognizes that the audit is already authoritative, contains the observer failure, and completes the pair. The one precondition: ask from inside an open turn — the audit pair is turn-enclosed by contract (the turn is the durable log's commit/replay boundary; a bare event between turns is crash-tail garbage on reload), so an idle ask throws before appending anything.
The contract in one line: `ctx.approval.request(req)` puts exactly one question — "may this specific action proceed?" — to whatever answerers the deployment composed, and its answerer phase always produces an outcome: an aborted signal yields `cancelled`, a throwing or missing answerer yields `unavailable`, and `allowed-once` is a grant for the single asked-about action, never a class of future ones. `ApprovalRequest` is a readonly same-process contract: the service borrows the exact request, agent, session, and abort signal rather than cloning or freezing them. The request requires an open turn because the audit pair is turn-enclosed by contract (the turn is the durable log's commit/replay boundary; a bare event between turns is crash-tail garbage on reload), so an idle ask rejects before appending. Either audit append may reject before commit because returning an unlogged decision would violate the pair. Session contains post-commit observer failures, so an authoritative audit append cannot reject the request or suppress its matching event.
The service is the mechanism, answerers are the policy. Answerers are `approval/request` waterfall listeners occupying a single decision slot: answer for an agent you own by returning an outcome without calling `next()`, or delegate an agent you don't recognize by calling `next()` — the chain's built-in default is `unavailable`, so a deployment with no answerer (headless, CI) fails closed with zero configuration. Dispatch is keyed by `req.agent`: a listener registered through `agent.ctx` receives only that agent's questions, while a plain-context listener receives every agent's. Registration order across sibling plugins is not load-order deterministic; compose one terminal answerer per deployment and use `prepend` listeners only for decide-or-delegate gates.
The seam also owns the per-session POLICY tier ([the sandbox RFC § Per-session mode switching](../../../docs/rfc/implemented/feature/2026-07-06-sandbox.md)): `ApprovalPolicy` is `'ask'` (delegate to the answerers) or `'never'` (deterministically reject without prompting anyone; the strict CI/unattended stance), with `effective = fold(the session's 'approval/policy' events, last one wins) ?? Config.policy` — the session log is the store, written only through `setApprovalPolicy(session, policy)`. The service decides `'never'` inside `request()` itself, before dispatching the waterfall (`'never'` → `'rejected'` with the audit pair still landing; no listener registration, including a later `prepend`, can precede it), states `'never'` — and only `'never'` in prose — in a per-agent prompt section, records either value with a source-owned header marker, and narrates a policy switch to the model in at most one coalesced `agent/pre-step` notice. The restart fallback reads the marker rather than deployment-controlled persona prose; attribution is positional (an override event after the last `request/header*` reads `changed by the user`, otherwise `changed by the operator/config`).
The seam also owns the per-session POLICY tier ([the sandbox RFC § Per-session mode switching](../../../docs/rfc/implemented/feature/2026-07-06-sandbox.md)): `ApprovalPolicy` is `'ask'` (delegate to the answerers) or `'never'` (deterministically reject without prompting anyone; the strict CI/unattended stance), with `effective = fold(the session's 'approval/policy' events, last one wins) ?? Config.policy` — the session log is the store, written only through `setApprovalPolicy(session, policy)`, which rejects any value outside that closed vocabulary before appending. The service decides `'never'` inside `request()` itself, before dispatching the waterfall (`'never'` → `'rejected'` with the audit pair still landing; no listener registration, including a later `prepend`, can precede it), states `'never'` — and only `'never'` in prose — in a per-agent prompt section, records either value with a source-owned header marker, and narrates a policy switch to the model in at most one coalesced `agent/pre-step` notice. The restart fallback reads the marker rather than deployment-controlled persona prose; attribution is positional (an override event after the last `request/header*` reads `changed by the user`, otherwise `changed by the operator/config`).
One seam serves both ask paths of [the sandbox RFC](../../../docs/rfc/implemented/feature/2026-07-06-sandbox.md): the `tools/pre-execute` `ask` decision (routed by [`@deepseek-ai/dsh-tools`](../../core/tools/) when this service is mounted; degrading to deny when it is not), and the sandbox post-denial escalated retry (the bash tool's `sandbox_permissions` gate in [`@deepseek-ai/dsh-tool-bash`](../../bash/tool-bash/) — [the sandbox RFC § Escalation](../../../docs/rfc/implemented/feature/2026-07-06-sandbox.md)). The full design: [the approval-seam RFC](../../../docs/rfc/implemented/feature/2026-07-06-approval-seam.md).

154
packages/ui/user-approval/src/index.ts
View File

@@ -63,10 +63,8 @@ declare module 'cordis' {
* Scope-filtered dispatch (`@deepseek-ai/dsh-scope`) keys the carrier by `req.agent`: a
* listener registered through `agent.ctx` receives only that agent's
* questions, while a plain-context listener receives every agent's.
* `req` is the service's shallow-frozen acceptance snapshot: later caller
* mutation cannot redirect the question, while the `agent` and `signal`
* identity capabilities remain exact.
* @param req - the accepted decision (agent, tool identity, reason, signal).
* `req` is a readonly same-process value borrowed from the caller.
* @param req - the pending decision (agent, tool identity, reason, signal).
* @mode waterfall
*/
'approval/request'(this: Scoped<ApprovalService>, req: ApprovalRequest, next: () => Promise<ApprovalOutcome>): Promise<ApprovalOutcome>
@@ -223,12 +221,16 @@ function hasOpenTurn(events: readonly SessionEvent[]): boolean {
* THE write path for a session's approval-policy override: appends exactly
* one `approval/policy` event — the switch IS its event; nothing mutates
* policy state out of band. Takes effect on the session's next ask and next
* prompt assembly (the consumers fold on every read).
* prompt assembly (the consumers fold on every read). Rejects a value outside
* {@link APPROVAL_POLICIES} before appending anything.
* @param session - the session the override belongs to.
* @param policy - the policy every subsequent ask for this session resolves
* under (until the next switch).
*/
export function setApprovalPolicy(session: Session, policy: ApprovalPolicy): void {
if (!APPROVAL_POLICIES.includes(policy)) {
throw new TypeError('approval policy must be one of "ask" or "never"')
}
session.append('approval/policy', { policy })
}
@@ -237,9 +239,9 @@ export function setApprovalPolicy(session: Session, policy: ApprovalPolicy): voi
* for an answerer to present it and for the audit events to reconstruct what
* was asked — it deliberately does NOT carry tool arguments: a UI answerer
* attaches the prompt to the already-streamed tool call via `callId` instead
* of re-rendering the call. `request()` synchronously copies and shallow-freezes
* this record before crossing an asynchronous boundary. Scalar fields are
* detached; the `agent` and `signal` identity capabilities are preserved.
* of re-rendering the call. This is a readonly same-process contract:
* `request()` borrows the request and its `agent` and `signal` capabilities
* directly rather than treating them as serialized input.
*/
export interface ApprovalRequest {
/**
@@ -247,21 +249,21 @@ export interface ApprovalRequest {
* UI answerer only answers for agents it owns) and receives the audit
* events on its session log.
*/
agent: Agent
readonly agent: Agent
/** The tool the question is about (presentation and audit). */
toolName: string
readonly toolName: string
/**
* The exact tool call being decided, when the asker has one — lets a UI
* attach the prompt to the tool call it already streamed.
*/
callId?: CallId
readonly callId?: CallId
/** The asker's human-readable explanation of WHY it is asking. */
reason?: string
readonly reason?: string
/**
* Aborting withdraws the question: the request settles `'cancelled'`
* immediately and a late answer from a still-pending answerer is discarded.
*/
signal?: AbortSignal
readonly signal?: AbortSignal
}
/** Plugin config. All optional — `static Config` supplies the defaults. */
@@ -272,7 +274,7 @@ export interface Config {
* (fail-closed with none); `'never'` auto-rejects every ask without
* prompting (the deterministic CI/unattended stance).
*/
policy?: ApprovalPolicy
readonly policy?: ApprovalPolicy
}
/**
@@ -297,7 +299,7 @@ export class ApprovalService extends Service {
constructor(ctx: Context, public config: Config) {
super(ctx, 'approval')
const effective = (agent: Agent): ApprovalPolicy => this.effectivePolicy(agent)
const effective = (agent: Agent): ApprovalPolicy => this.effectivePolicy(agent.session)
// Visibility layer 1, scoped on the prompt registry so headless
// compositions mount the seam without it: state the one deterministic
@@ -345,7 +347,7 @@ export class ApprovalService extends Service {
}
// Same fold effectivePolicy performs — override is scanned here anyway
// for POSITIONAL attribution; the default lives once, in the method.
const current = this.effectivePolicy(agent)
const current = this.effectivePolicy(session)
const header = session.requestHeader()
const told = narrated.get(session) ?? toldApprovalPolicy(header?.system)
narrated.set(session, current)
@@ -361,43 +363,25 @@ export class ApprovalService extends Service {
}
/**
* Ask the composed answerers to decide one request. Requires an open turn
* on the requesting agent's session — the audit pair below is turn-enclosed
* by contract (the turn is the log's commit/replay boundary; an idle append
* would be dropped as crash tail on reload) — and throws before appending
* anything when called idle; asking outside a turn is a deferred design.
* Within that precondition it always resolves to an outcome, never rejects:
* an aborted signal yields `'cancelled'`, a missing or throwing answerer
* yields `'unavailable'` (fail closed), and a rogue non-vocabulary return
* value is normalized to `'unavailable'`. The caller-owned request is
* synchronously snapshotted, so later mutation cannot split routing,
* dispatch payload, cancellation, or the audit pair across agents/sessions.
* Appends the
* `approval/asked`/`approval/decided` audit pair (log-only) around the
* decision regardless of outcome. A synchronous session observer failure
* after an audit event entered the append-only log is contained; the event
* is already authoritative, so the pair still completes and the request
* still resolves.
* Ask the composed answerers to decide one readonly same-process request.
* The service borrows the request, agent, session, and live signal directly.
* The request requires an open turn because the audit pair must be enclosed
* by the durable log's commit/replay boundary; an idle ask rejects before
* appending anything. The answerer phase always produces an outcome: an
* aborted signal yields `'cancelled'`, a missing or throwing answerer yields
* `'unavailable'` (fail closed), and a rogue non-vocabulary return value is
* normalized to `'unavailable'`. A failure that prevents either audit append
* from committing still rejects because returning an unlogged decision would
* violate the pair. Session contains post-commit observer failures, so an
* authoritative append cannot reject the request or suppress its matching
* audit event.
* @param req - the pending decision (agent, tool identity, reason, signal).
* @returns the closed outcome; `'allowed-once'` is the only grant.
* @throws when no turn is open or either audit event fails before the session
* append commit point.
*/
async request(req: ApprovalRequest): Promise<ApprovalOutcome> {
// Accept one immutable request shape before the first async boundary. The
// caller retains its record and may mutate it as soon as this async method
// returns; identity capabilities stay live, but the record is never reread.
const agent = req.agent
const toolName = req.toolName
const callId = req.callId
const reason = req.reason
const signal = req.signal
const accepted: Readonly<ApprovalRequest> = Object.freeze({
agent,
toolName,
...callId !== undefined ? { callId } : {},
...reason !== undefined ? { reason } : {},
...signal !== undefined ? { signal } : {},
})
const session = accepted.agent.session
const session = req.agent.session
if (!hasOpenTurn(session.events)) {
throw new Error(
'approval.request() outside an open turn: the approval/asked + approval/decided audit pair '
@@ -406,67 +390,43 @@ export class ApprovalService extends Service {
)
}
const id = ApprovalRequestId(randomUUID())
this.appendAudit(session, 'approval/asked', id, () => {
session.append('approval/asked', {
id,
toolName: accepted.toolName,
...accepted.callId !== undefined ? { callId: accepted.callId } : {},
...accepted.reason !== undefined ? { reason: accepted.reason } : {},
})
})
const outcome = await this.decide(accepted)
this.appendAudit(session, 'approval/decided', id, () => {
session.append('approval/decided', { id, outcome })
session.append('approval/asked', {
id,
toolName: req.toolName,
...req.callId !== undefined ? { callId: req.callId } : {},
...req.reason !== undefined ? { reason: req.reason } : {},
})
const outcome = await this.decide(req, session)
session.append('approval/decided', { id, outcome })
return outcome
}
/**
* Append one audit event while distinguishing a post-append observer throw
* from a failure that prevented the event entering the log. `Session.append`
* pushes first and then notifies synchronously, so log growth proves the
* event is already authoritative; that observer failure is reported and
* contained so it cannot reject the approval or suppress its matching event.
* @param session - the captured session receiving both audit events.
* @param type - the audit event currently being appended.
* @param id - the request id, used to identify the contained failure.
* @param append - the single concrete `Session.append` call.
*/
private appendAudit(
session: Session,
type: 'approval/asked' | 'approval/decided',
id: ApprovalRequestId,
append: () => void,
): void {
const length = session.events.length
try {
append()
} catch (error) {
if (session.events.length === length) throw error
this.ctx.logger.warn(`approval request "${id}": ${type} observer threw after the event was appended`)
}
}
/**
* The session's effective policy: its own `approval/policy` fold, else the
* configured default (the schema already defaulted an omitted policy to
* `'ask'`; the `??` only narrows the optional-input TYPE).
* @param agent - the agent whose session's policy applies.
* @returns the policy every ask for this agent resolves under right now.
* @param session - the exact accepted session whose policy applies.
* @returns the policy every ask for this session resolves under right now.
*/
private effectivePolicy(agent: Agent): ApprovalPolicy {
return effectiveApprovalPolicy(agent.session.events) ?? this.config.policy ?? 'ask'
private effectivePolicy(session: Session): ApprovalPolicy {
return effectiveApprovalPolicy(session.events) ?? this.config.policy ?? 'ask'
}
/** Dispatch the waterfall, contained and raced against the accepted signal. */
private async decide(req: Readonly<ApprovalRequest>): Promise<ApprovalOutcome> {
if (req.signal?.aborted) return 'cancelled'
/**
* Dispatch the waterfall, contained and raced against the request signal.
* @param req - the borrowed public request.
* @param session - the request agent's session used for policy lookup.
* @returns the normalized closed outcome.
*/
private async decide(req: ApprovalRequest, session: Session): Promise<ApprovalOutcome> {
const signal = req.signal
if (signal?.aborted) return 'cancelled'
// The 'never' policy is decided HERE, before any dispatch: a listener
// registered with `prepend: true` after this service mounts would sit
// ahead of any gate LISTENER, so a listener-shaped gate cannot keep the
// documented promise that 'never' rejects deterministically regardless
// of registration order — only the service's own request path can.
if (this.effectivePolicy(req.agent) === 'never') return 'rejected'
if (this.effectivePolicy(session) === 'never') return 'rejected'
// Enter the promise chain BEFORE dispatching: a listener that throws
// SYNCHRONOUSLY (before its first await) must land in the same rejection
// path as an async one — `Promise.resolve(call())` would let it escape
@@ -484,10 +444,12 @@ export class ApprovalService extends Service {
// tool call open — the seam contains its callbacks.
() => 'unavailable',
)
const signal = req.signal
if (signal === undefined) return answer
return await new Promise<ApprovalOutcome>((resolve) => {
const onAbort = () => { resolve('cancelled') }
const onAbort = () => {
signal.removeEventListener('abort', onAbort)
resolve('cancelled')
}
signal.addEventListener('abort', onAbort, { once: true })
void answer.then((outcome) => {
signal.removeEventListener('abort', onAbort)

122
packages/ui/user-approval/tests/approval.spec.ts
View File

@@ -2,7 +2,8 @@ import { describe, expect, it, vi } from 'vitest'
import { Context } from 'cordis'
import type { Agent } from '@deepseek-ai/dsh-agent'
import { CallId } from '@deepseek-ai/dsh-llm'
import { carrierKeyOf, scopeHost } from '@deepseek-ai/dsh-scope'
import { carrierKeyOf, createScope } from '@deepseek-ai/dsh-scope'
import type { Scope } from '@deepseek-ai/dsh-scope'
import SessionStore, { Session, SessionId } from '@deepseek-ai/dsh-session'
import type { SessionEvent } from '@deepseek-ai/dsh-session'
import SystemPrompt from '@deepseek-ai/dsh-system-prompt'
@@ -78,77 +79,38 @@ describe('ApprovalService.request', () => {
expect(Object.keys(appended[0]?.data ?? {}).sort()).toEqual(['id', 'toolName'])
})
it('snapshots request identity, scope, payload, and audit before deferred dispatch', async () => {
it('borrows the exact readonly request for scoped dispatch and audit', async () => {
const ctx = await mounted()
const { agent: acceptedAgent, appended: acceptedAudit } = fakeAgent()
const { agent: replacementAgent, appended: replacementAudit } = fakeAgent()
const host = await scopeHost(ctx, ['approval'])
const acceptedScope = host.mint(acceptedAgent)
const replacementScope = host.mint(replacementAgent)
const dispatchStarted = Promise.withResolvers<'started'>()
const answer = Promise.withResolvers<ApprovalOutcome>()
const originalSignal = new AbortController().signal
const replacementSignal = new AbortController().signal
let heardBy: 'accepted' | 'replacement' | undefined
const { agent, appended } = fakeAgent()
let scope!: Scope
const scopeFiber = await ctx.plugin(Object.assign((inner: Context) => {
scope = createScope(inner, agent)
}, { inject: ['approval'] }))
let received: ApprovalRequest | undefined
let carrier: unknown
acceptedScope.ctx.on('approval/request', function (req) {
heardBy = 'accepted'
scope.ctx.on('approval/request', function (req) {
received = req
carrier = carrierKeyOf(this)
dispatchStarted.resolve('started')
return answer.promise
return Promise.resolve<ApprovalOutcome>('allowed-once')
})
replacementScope.ctx.on('approval/request', function (req) {
heardBy = 'replacement'
received = req
carrier = carrierKeyOf(this)
dispatchStarted.resolve('started')
return answer.promise
})
const request = requestOf(acceptedAgent, {
toolName: 'original-tool',
callId: CallId('original-call'),
reason: 'original reason',
signal: originalSignal,
const request = requestOf(agent, {
toolName: 'scoped-tool',
callId: CallId('scoped-call'),
reason: 'scoped reason',
})
const pending = ctx.approval.request(request)
// request() has returned, but the answerer dispatch is deliberately queued
// in a microtask. Mutating the caller-owned record must not redirect it.
request.agent = replacementAgent
request.toolName = 'mutated-before-dispatch'
request.callId = CallId('mutated-call')
request.reason = 'mutated reason'
request.signal = replacementSignal
await dispatchStarted.promise
// Mutation while the answer is pending must not redirect the final audit.
request.toolName = 'mutated-after-dispatch'
request.reason = 'mutated again'
answer.resolve('allowed-once')
await expect(pending).resolves.toBe('allowed-once')
expect(heardBy).toBe('accepted')
expect(carrier).toBe(acceptedAgent)
expect(received).not.toBe(request)
expect(Object.isFrozen(received)).toBe(true)
expect(received).toMatchObject({
agent: acceptedAgent,
toolName: 'original-tool',
callId: 'original-call',
reason: 'original reason',
signal: originalSignal,
await expect(ctx.approval.request(request)).resolves.toBe('allowed-once')
expect(carrier).toBe(agent)
expect(received).toBe(request)
expect(appended).toHaveLength(2)
expect(appended[0]?.data).toMatchObject({
toolName: 'scoped-tool',
callId: 'scoped-call',
reason: 'scoped reason',
})
expect(acceptedAudit).toHaveLength(2)
expect(acceptedAudit[0]?.data).toMatchObject({
toolName: 'original-tool',
callId: 'original-call',
reason: 'original reason',
})
expect(acceptedAudit[1]?.data).toMatchObject({ outcome: 'allowed-once' })
expect(acceptedAudit[1]?.data['id']).toBe(acceptedAudit[0]?.data['id'])
expect(replacementAudit).toEqual([])
await host.dispose()
expect(appended[1]?.data).toMatchObject({ outcome: 'allowed-once' })
expect(appended[1]?.data['id']).toBe(appended[0]?.data['id'])
await scopeFiber.dispose()
})
it('contains an approval/asked observer throw after append and still completes the pair', async () => {
@@ -171,7 +133,7 @@ describe('ApprovalService.request', () => {
const decided = session.events.find((event): event is SessionEvent<'approval/decided'> => event.type === 'approval/decided')
expect(audit.map(event => event.type)).toEqual(['approval/asked', 'approval/decided'])
expect(decided?.data.id).toBe(asked?.data.id)
expect(warn).toHaveBeenCalledWith(expect.stringContaining('approval/asked observer threw'))
expect(warn).toHaveBeenCalledWith(expect.stringContaining('session/event listener threw: Error: observer failed after asked append'))
})
it('contains an approval/decided observer throw after append and still resolves', async () => {
@@ -194,10 +156,10 @@ describe('ApprovalService.request', () => {
const decided = session.events.find((event): event is SessionEvent<'approval/decided'> => event.type === 'approval/decided')
expect(audit.map(event => event.type)).toEqual(['approval/asked', 'approval/decided'])
expect(decided?.data).toMatchObject({ id: asked?.data.id, outcome: 'rejected' })
expect(warn).toHaveBeenCalledWith(expect.stringContaining('approval/decided observer threw'))
expect(warn).toHaveBeenCalledWith(expect.stringContaining('session/event listener threw: Error: observer failed after decided append'))
})
it('does not misclassify a pre-append failure as an observer failure', async () => {
it('propagates an append failure that prevented audit log growth', async () => {
const ctx = await mounted()
const failure = new Error('append failed before log growth')
const agent = {
@@ -236,9 +198,12 @@ describe('ApprovalService.request', () => {
const ctx = await mounted()
const { agent: agentA } = fakeAgent()
const { agent: agentB } = fakeAgent()
const host = await scopeHost(ctx, ['approval'])
const scopeA = host.mint(agentA)
const scopeB = host.mint(agentB)
let scopeA!: Scope
let scopeB!: Scope
const scopesFiber = await ctx.plugin(Object.assign((inner: Context) => {
scopeA = createScope(inner, agentA)
scopeB = createScope(inner, agentB)
}, { inject: ['approval'] }))
const heard: string[] = []
ctx.on('approval/request', (req, next) => {
heard.push(req.agent === agentA ? 'global:A' : 'global:B')
@@ -257,14 +222,16 @@ describe('ApprovalService.request', () => {
await expect(ctx.approval.request(requestOf(agentB))).resolves.toBe('unavailable')
expect(heard).toEqual(['global:A', 'scoped:A', 'global:B', 'scoped:B'])
await host.dispose()
await scopesFiber.dispose()
})
it('keys the scoped dispatch carrier to the exact request agent', async () => {
const ctx = await mounted()
const { agent } = fakeAgent()
const host = await scopeHost(ctx, ['approval'])
const scope = host.mint(agent)
let scope!: Scope
const scopeFiber = await ctx.plugin(Object.assign((inner: Context) => {
scope = createScope(inner, agent)
}, { inject: ['approval'] }))
let seenKey: object | undefined
scope.ctx.on('approval/request', function (req, next) {
seenKey = carrierKeyOf(this)
@@ -275,7 +242,7 @@ describe('ApprovalService.request', () => {
await expect(ctx.approval.request(requestOf(agent))).resolves.toBe('unavailable')
expect(seenKey).toBe(agent)
await host.dispose()
await scopeFiber.dispose()
})
it('contains a throwing answerer as unavailable', async () => {
@@ -420,6 +387,15 @@ describe('approval policy (the approval/policy fold)', () => {
expect(session.events.at(-1)).toMatchObject({ type: 'approval/policy', data: { policy: 'ask' } })
})
it('rejects a policy outside the closed vocabulary before appending', () => {
const append = vi.fn()
const session = { append } as unknown as Session
expect(() => { setApprovalPolicy(session, 'sometimes' as Parameters<typeof setApprovalPolicy>[1]) })
.toThrow('approval policy must be one of "ask" or "never"')
expect(append).not.toHaveBeenCalled()
})
it('defaults a schema-less construction to ask (the ?? narrows the optional TYPE)', async () => {
// Direct construction bypasses the plugin schema (the SystemPrompt-test
// precedent for covering a defaulted Config field's type-narrowing ??).