openclaw ACP adapter: cancellation marker survives a finished prompt and mislabels the next prompt as cancelled (#3997)

Finding (from PR review)

A cancellation that races with a prompt's completion leaves a stale cancellation marker that mislabels the next normally completed prompt. Codex (P2, PR #3997):

When a cancellation notification races with the prompt's final idle event, the prompt can finish before this handler runs. The handler still records the session as cancelled, but an ensuing idle event cannot remove that marker because cleanup in handleServeEvent requires a pending prompt. The next normally completed prompt then incorrectly returns stopReason: "cancelled". The marker also survives a cancelled prompt ending through session.error. Only mark active prompts as cancelled and clear their cancellation state on every terminal path.

Evidence

Source PR: feat(openclaw): ACP-to-HTTP adapter so OpenClaw shares the standing opencode serve (#1243) #3997 — feat(openclaw): ACP-to-HTTP adapter so OpenClaw shares the standing opencode serve (#1243) #3997 (merged 2026-11-02 as-is)
Review comment: feat(openclaw): ACP-to-HTTP adapter so OpenClaw shares the standing opencode serve (#1243) #3997 (comment) — chatgpt-codex-connector[bot], 2026-11-01T21:31:19Z, bin/opencode-acp-adapter.mjs:422
No-response evidence: the review thread is unresolved (GraphQL reviewThreads, checked 2026-11-02), the comment has no replies, and PR feat(openclaw): ACP-to-HTTP adapter so OpenClaw shares the standing opencode serve (#1243) #3997 contains a single commit (8e67215, 2026-11-01T21:26:51Z) that predates the comment — no fix followed it before merge.
Still live on current main (verified by raw read 2026-11-02): bin/opencode-acp-adapter.mjs:422 — session/cancel calls cancelledSessions.add(sessionId) unconditionally; the marker is deleted only at :308-312, inside the session.idle branch gated on pendingPrompts.has(sessionId), and the session.error path (finishPrompt) never clears it.

Acceptance criteria

Cancellation is recorded only for prompts actually in flight, and cancellation state is cleared on every terminal path (session.idle, session.error, prompt finish).
A regression test covers: cancel arriving after a prompt finished, followed by a normal prompt that must return stopReason: "end_turn".
Or the finding is resolved by a documented decision on this issue.

🤖 Agent note · filed

Agent: pr-comment-review job
Session / ID: pr-comment-review cron run 2026-11-02 15:54 CST (2026-11-02T22:14:31Z)
Found via: unaddressed PR comment — PR feat(openclaw): ACP-to-HTTP adapter so OpenClaw shares the standing opencode serve (#1243) #3997, Codex comment feat(openclaw): ACP-to-HTTP adapter so OpenClaw shares the standing opencode serve (#1243) #3997 (comment)
Task: Triage the validated finding: cancellation marker survives a finished prompt (and the session.error path) and mislabels the next prompt as cancelled (thread unresolved, no reply, no post-comment fix commit, PR merged as-is, still live on current main)
Expectation: effort s — queue worker verifies against current main, then fixes or documents a decision
Links: issue openclaw ACP adapter: cancellation marker survives a finished prompt and mislabels the next prompt as cancelled (#3997) #4618 · source PR feat(openclaw): ACP-to-HTTP adapter so OpenClaw shares the standing opencode serve (#1243) #3997

🤖 Agent note · handed off

Agent: queue worker (repo-worker-code), lane code-h
Session / ID: task 6520
Found via: queue claim
Task: Handed off to the opencode execution lane for triage of issue openclaw ACP adapter: cancellation marker survives a finished prompt and mislabels the next prompt as cancelled (#3997) #4618 — the laptop claimant launches the bound opencode session, which drives task 6520 to its terminal state.
Expectation: s — verify the stale cancellation marker in bin/opencode-acp-adapter.mjs against current main, then fix-with-test or a documented decision per the packet.
Links: issue openclaw ACP adapter: cancellation marker survives a finished prompt and mislabels the next prompt as cancelled (#3997) #4618, PR feat(openclaw): ACP-to-HTTP adapter so OpenClaw shares the standing opencode serve (#1243) #3997

agent-note:v1
event: handed_off
state: handed off
task: 6520 (repo-health-code-pilot; worker repo-worker-code)
session: ses_e5e61a4e311ffea3y9BCxWDFUd
route: opencode-route/workhorse
action: Bound OpenCode worker launched for this already-claimed task; it must not claim another task.

🤖 Agent note · picked up

Agent: repo-worker-code (opencode lane)
Session / ID: ses_e5e61a4e311ffea3y9BCxWDFUd
Found via: already-claimed queue task 6520
Task: Verify issue openclaw ACP adapter: cancellation marker survives a finished prompt and mislabels the next prompt as cancelled (#3997) #4618 against current main and determine whether stale cancellation markers can misreport the next prompt; implement the narrow fix and regression test if confirmed.
Expectation: s — fix with a targeted regression test, or document a decision on this issue.
Links: issue openclaw ACP adapter: cancellation marker survives a finished prompt and mislabels the next prompt as cancelled (#3997) #4618 · source PR feat(openclaw): ACP-to-HTTP adapter so OpenClaw shares the standing opencode serve (#1243) #3997

🤖 Agent note · parked

Agent: repo-worker-code (opencode lane)
Session / ID: ses_e5e61a4e311ffea3y9BCxWDFUd
Found via: queue task 6520; fresh GitHub main read
Task: Confirmed the defect remains on current main. session/cancel adds to cancelledSessions unconditionally (lines 419–426); session.idle clears it only while a prompt is pending (307–312); session.error finishes the prompt without clearing it (314–320).
Expectation: s — resume for the narrow fix and regression test. No code decision waives the defect.
Blocker / resume condition: The runbook requires edits and tests on the laptop. The approved laptop SSH helper could not start because this session has no HTTPS_PROXY runtime environment; no laptop worktree was created and no sandbox edits/tests were made. Resume when the approved laptop SSH runtime is restored.
Links: issue openclaw ACP adapter: cancellation marker survives a finished prompt and mislabels the next prompt as cancelled (#3997) #4618 · PR feat(openclaw): ACP-to-HTTP adapter so OpenClaw shares the standing opencode serve (#1243) #3997 · current source: https://github.com/developing-today/code/blob/main/bin/opencode-acp-adapter.mjs#L307-L320

🤖 Agent note · handed off

Agent: fleet monitor
Session / ID: repo-fleet-monitor run 2026-11-02 20:12 CST
Found via: monitor reconciliation
Task: Claim-evaporated repair: task 6520 is queued unclaimed (attempts 3/3, checkpoint stage handed_off) while this issue still carried claimed labels. Labels reset to queue/queued + status/needs-triage so GitHub matches the queue. The task is DEAD at its attempt cap and can never be claimed again — operator repair: raise the cap or enqueue a fresh packet resuming from stage handed_off.
Expectation: operator/lane pickup after repair; no holder is currently working this issue.

🤖 Agent note · handed off

Agent: queue worker (repo-worker-code), lane code-h
Session / ID: task 6520
Found via: queue claim
Task: Re-handing off task 6520 (triage, issue openclaw ACP adapter: cancellation marker survives a finished prompt and mislabels the next prompt as cancelled (#3997) #4618 — stale ACP cancellation marker) to the opencode execution lane. Verified fresh via REST: issue openclaw ACP adapter: cancellation marker survives a finished prompt and mislabels the next prompt as cancelled (#3997) #4618 is OPEN with no hard-stop labels. Prior progress on task 6520 names PR fix(acp): clear stale cancellation markers on every terminal path (#4618) #4642 OPEN (basic guard SUCCESS at the last check) on bin/opencode-acp-adapter.mjs — the claimant session should resume from PR fix(acp): clear stale cancellation markers on every terminal path (#4618) #4642, not restart.
Expectation: s/m — claimant session drives task 6520 to a terminal state, resuming PR fix(acp): clear stale cancellation markers on every terminal path (#4618) #4642.
Links: issue openclaw ACP adapter: cancellation marker survives a finished prompt and mislabels the next prompt as cancelled (#3997) #4618; PR fix(acp): clear stale cancellation markers on every terminal path (#4618) #4642

🤖 Agent note · bound

Agent: fleet monitor (repo-fleet-monitor)
Session / ID: task 6520 · fleet-monitor run 2026-11-03 04:12 CST
Found via: monitor reconciliation (Duty 1B PR↔queue binding)
Task: Verified open PR fix(acp): clear stale cancellation markers on every terminal path (#4618) #4642 implements this issue (title/body/head checked against the issue). Task 6520 was DEAD at its attempts cap; the monitor's dead-in-queue self-repair raised the cap, so task 6520 is claimable again — the claiming lane should claim it, checkpoint stage pr_opened (artifact: PR fix(acp): clear stale cancellation markers on every terminal path (#4618) #4642), and park it as the PR's shepherd (review CI, merge when green, do NOT re-implement). Labels left as queue/claimed + work/assigned because PR work is in flight; the shepherd claim makes that true. (This note was re-posted via the REST API after the run's GraphQL attempts hit the fleet-wide rate limit.)
Expectation: PR fix(acp): clear stale cancellation markers on every terminal path (#4618) #4642 shepherded to merge; issue closes with it.

🤖 Agent note · picked up

Agent: queue worker (repo-worker-code), lane aea12c82 (openchamber scheduled lane)
Session / ID: task 6520
Found via: queue claim
Task: Resumed task 6520 from checkpoint stage handed_off. Verified fresh: issue OPEN, no hard-stop labels, PR fix(acp): clear stale cancellation markers on every terminal path (#4618) #4642 OPEN and owned by task 6520. PR was CONFLICTING with main, so merged origin/main into queue/6520-4618-stale-cancel-marker — kept the openclaw ACP adapter: cancellation marker survives a finished prompt and mislabels the next prompt as cancelled (#3997) #4618 guard (clear the cancel marker on every terminal path, ignore cancel when no prompt is pending), adopted main's openclaw ACP adapter: load the standing-server credential — gateway launch wires no OPENCODE_SERVER_PASSWORD (#3997) #4615 auth tests and data.error message shape, dropped partTextSent.clear() (variable removed on main). Head now f9fd9ff.
Expectation: targeted test node --test bin/opencode-acp-adapter.test.mjs passes 9/9; merge fix(acp): clear stale cancellation markers on every terminal path (#4618) #4642 when the PR tier is green.
Links: PR fix(acp): clear stale cancellation markers on every terminal path (#4618) #4642; head f9fd9ff

Merge pull request #4642 from developing-today/queue/6520-4618-stale-…

🤖 Agent note · done

Agent: queue worker (repo-worker-code), lane aea12c82 (openchamber scheduled lane)
Session / ID: task 6520
Found via: queue claim (resumed from handed_off)
Outcome: PR fix(acp): clear stale cancellation markers on every terminal path (#4618) #4642 merged at 2026-11-03T12:06:51Z as 4372e61, head f9fd9ff. Cancel marker is now cleared on every terminal path (idle/error/submission failure) and session/cancel with no pending prompt no longer arms a stale marker, so the next prompt keeps its own end_turn. Merge resolved main's openclaw ACP adapter: load the standing-server credential — gateway launch wires no OPENCODE_SERVER_PASSWORD (#3997) #4615 auth-test additions alongside the openclaw ACP adapter: cancellation marker survives a finished prompt and mislabels the next prompt as cancelled (#3997) #4618 regression test.
Verified: node --test bin/opencode-acp-adapter.test.mjs — 9/9 pass locally on the merged head; PR tier basic: OpenCode metered-provider guard pass.
Expectation: issue closed by the merge; no follow-up work needed.
Links: issue openclaw ACP adapter: cancellation marker survives a finished prompt and mislabels the next prompt as cancelled (#3997) #4618; PR fix(acp): clear stale cancellation markers on every terminal path (#4618) #4642; merge commit 4372e61

Navigation Menu

Repository navigation

openclaw ACP adapter: cancellation marker survives a finished prompt and mislabels the next prompt as cancelled (#3997) #4618

Finding (from PR review)

Evidence

Acceptance criteria

32 remaining items

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

openclaw ACP adapter: cancellation marker survives a finished prompt and mislabels the next prompt as cancelled (#3997) #4618

Description

Finding (from PR review)

Evidence

Acceptance criteria

Activity

dezren39 commented on Nov 2, 2026

dezren39 commented on Nov 2, 2026

dezren39 commented on Nov 2, 2026

dezren39 commented on Nov 2, 2026

dezren39 commented on Nov 2, 2026

32 remaining items

dezren39 commented on Nov 3, 2026

dezren39 commented on Nov 3, 2026

dezren39 commented on Nov 3, 2026

dezren39 commented on Nov 3, 2026

dezren39 commented on Nov 3, 2026

Metadata

Metadata

Assignees

Labels

Type

Projects

Milestone

Relationships

Development

Issue actions