Support multiple work-queue claims and claim-scoped safe outputs by Copilot · Pull Request #65682 · github/gh-aw

Copilot · 2026-10-04T23:38:53Z

Workers can receive multiple claims in aw_context, but the existing finish and safe-output flow assumes one claim. This change supports partial completion without applying outputs for unfinished claims.

Claim lifecycle: Accept a single assignment or an array. Array assignments require a finish intent for every claim_id; a missing intent cancels the batch and blocks safe outputs.
Safe outputs: Tool calls may include claim_id. Tagged outputs apply only after that claim’s completion is verified. Untagged outputs and custom steps/actions require every claim to complete.
Compatibility: Preserve single-claim behavior and update worker prompts, documentation, and coverage.

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>

Copilot

Copilot review overview

🟡 Changes recommended

Some side-effect jobs bypass claim reconciliation, and an existing integration assertion is now stale.

Review effort: Balanced
Findings: 2

Open (2)

Side-effect jobs bypass reconciliation and consume unfiltered artifacts · New
Update integration test for all_authorized user-step behavior · New

What changed in this PR

Adds multi-claim work-queue reconciliation and claim-scoped safe-output processing while retaining single-claim compatibility.

Changes:

Accepts and validates arrays of worker assignments and finish intents.
Filters safe outputs by successfully completed claims.
Updates compiler gating, prompts, documentation, and tests.

File	Description
`pkg/workflow/compiler_safe_outputs_job.go`	Adds all-claim gating for custom effects.
`pkg/workflow/compiler_safe_outputs_job_test.go`	Tests partial-claim step gating.
`actions/setup/js/write_work_queue_snapshot.cjs`	Admits multiple effective assignments.
`actions/setup/js/write_work_queue_snapshot.test.cjs`	Tests batch assignment validation.
`actions/setup/js/work_queue_summary.cjs`	Summarizes every assigned claim.
`actions/setup/js/work_queue_mcp_server.cjs`	Adds per-claim finish intents.
`actions/setup/js/work_queue_mcp_server.test.cjs`	Tests batch finish handling.
`actions/setup/js/generate_safe_outputs_tools.cjs`	Adds optional claim identifiers to tools.
`actions/setup/js/generate_safe_outputs_tools.test.cjs`	Tests claim-aware tool schemas.
`actions/setup/js/finish_work_queue_claim.cjs`	Reconciles batches and filters outputs.
`actions/setup/js/finish_work_queue_claim.test.cjs`	Covers partial and incomplete batches.
`actions/setup/js/create_prompt.cjs`	Instructs workers about multiple claims.
`actions/setup/js/collect_ndjson_output.cjs`	Preserves validated claim identifiers.
`actions/setup/js/aw_context.cjs`	Parses single or array assignments.
`.github/workflows/smoke-work-queue.lock.yml`	Regenerates the all-claims gate.
`.github/aw/work-queue.md`	Documents multi-claim behavior.

Copilot · 2026-10-05T05:10:54Z

+  const completed = new Set(workers.filter((_, index) => results[index].authorized).map(item => item.claim_id));
+  const allCompleted = completed.size === workers.length;
+  const authorized = completeBatch && completed.size > 0;
+  if (authorized) filterClaimOutputs(completed, allCompleted, options.agentOutputPath || `${TMP_GH_AW_PATH}/${AGENT_OUTPUT_FILENAME}`);


Fixed in f9a913a. Both downstream jobs now depend on safe_outputs and consume its filtered reconciliation artifact with no fallback to the original artifact. Asset uploads require claim authorization and reconciled output types; custom jobs additionally require all claims completed. Added partial/cancelled/completed batch tests and downstream compilation coverage. Focused queue integration, full workflow-package unit tests, and 68 JavaScript tests pass.

Copilot · 2026-10-05T05:10:55Z

 		}
 		if isWorkQueueEnabled(data) {
-			typedStep.If = combineGitHubIfExpressions(workQueueClaimAuthorizedExpression, typedStep.If)
+			typedStep.If = combineGitHubIfExpressions("steps.work_queue_claim_reconciliation.outputs.all_authorized == 'true'", typedStep.If)


Fixed in f9a913a: the user-step assertion now expects all_authorized, while ordinary handler assertions retain authorized. The queue compilation integration check passes locally.

gh-aw-bot · 2026-10-05T04:39:51Z

@copilot address the following outstanding work in one pass:

Update this branch with the latest main using make merge-main, resolving any conflicts and preserving the intended changes.
Review (actions/setup/js/finish_work_queue_claim.cjs:190): Filtering this job's local agent_output.json does not protect side-effect jobs that consume the original agent artifact. upload_assets is gated directly from needs.agent.outputs.output_types and depends on agent/activation (pkg/workflow/publish_assets.go:186-229), while custom safe jobs also depend on agent and download that unfiltered artifact (pkg/workflow/safe_jobs.go:335-414). Consequently, a missing/cancelled claim can still trigger these outputs (and untagged outputs can run during partial completion), bypassing the new reconciliation gate. Make every such job depend on r... - Support multiple work-queue claims and claim-scoped safe outputs #65682 (comment)
Review (pkg/workflow/compiler_safe_outputs_job.go:329): This changes user-provided steps to all_authorized, but TestWorkQueueCompilationPhases still requires the old outputs.authorized expression inside the user-step block (pkg/workflow/work_queue_compilation_integration_test.go:60-78). The integration suite will therefore fail for this intentional behavior change; update that assertion to expect outputs.all_authorized. - Support multiple work-queue claims and claim-scoped safe outputs #65682 (comment)
Fix failing check Compile queue phases (FAILURE): https://github.com/github/gh-aw/actions/runs/37244517597/job/111610812336.
Fix failing check impacted-go-tests (FAILURE): https://github.com/github/gh-aw/actions/runs/37244517712/job/111610886449.

Push the necessary fixes, reply to each listed review thread and resolve it when addressed. Ignore feedback already answered or resolved. Use the pr-finisher skill and stop when only human review or CI remains; do not trigger CI.

Sous-chef head: 128ee74
Sous-chef work: 1dcac6c4a72c1e9d2aa3c06d7b8bc0f55bbc03f204de26c5e88482aaab84b6f4 42e5d572d9d75b7168fe7d0aeda0717cc07e65ed240f6d299e6fec661214f1ac 57b160f058c0cf0ae8470606e768b98956d9e31df7d52cc35017efdeeaeffab1 8b1d0a9c7eb61f9904830caab0f89fed10be3caf4e6f080431ebf2d48580e274
Sous-chef state: f7cd44559e0f9fe51710d337806ae2b6a1dd40a2f89ae480eec6d094210fb6d3

Generated by 👨‍🍳 PR Sous Chef · pi · haiku45 · 3.82 AIC · ⌖ 8.31 AIC · ⊞ 3.2K · ◷
_{Comment /souschef to run again}

…ueue-api-support-multiple-claims Co-authored-by: gh-aw-bot <259018956+gh-aw-bot@users.noreply.github.com>

Co-authored-by: gh-aw-bot <259018956+gh-aw-bot@users.noreply.github.com>

github-actions · 2026-10-05T05:07:35Z

✅ Test Quality Sentinel completed test quality analysis.

Test Quality Sentinel skipped because pre-fetch PR data was unavailable: unable to fetch test file diff

🧪 Test quality analysis by Test Quality Sentinel

github-actions · 2026-10-05T05:07:35Z

🔎 PR Code Quality Reviewer is reviewing code quality for this pull request...

github-actions · 2026-10-05T05:07:37Z

✅ Design Decision Gate 🏗️ completed the design decision gate check. See the comment below for the result and any generated ADR draft.

🏗️ ADR gate enforced by Design Decision Gate 🏗️

github-actions · 2026-10-05T05:07:39Z

🧠 Matt Pocock Skills Reviewer has completed the skills-based review. ✅

🧠 Reviewed using Matt Pocock's skills by Matt Pocock Skills Reviewer

github-actions · 2026-10-05T05:07:48Z

✅ Ponytail Reviewer completed successfully!

Lean already. Ship.

Generated by Ponytail Reviewer for #65682

Copilot · 2026-10-05T05:10:55Z

@copilot address the following outstanding work in one pass:

Update this branch with the latest main using make merge-main, resolving any conflicts and preserving the intended changes....

Merged latest main using make merge-main (ded4e71) and pushed fixes in f9a913a. Closed the downstream artifact/authorization bypass, updated the integration assertion, fixed expression-valued concurrency queue schema validation, and corrected batch completion attempt collisions. Full workflow-package tests, queue integration, 68 JS tests, make lint, secret scanning, and 320-workflow drift validation pass. The progress gate still reports four pre-existing custom-linter findings in publish_assets.go. Replied to both review threads; thread resolution is unavailable with the authenticated tools here. Final automated review was unavailable and CodeQL timed out. CI on the new HEAD remains unverified; no CI was triggered.

github-actions · 2026-10-05T05:13:40Z

Comment Memory

Peek at saved memory (pr-code-quality-reviewer)

reviewed_at: 2026-10-05T05:08:59Z
review_event: REQUEST_CHANGES
submission_status: failed-to-submit-permission-denied
top_themes:
  - claim_id is exposed on unassigned work-queue runs
  - unknown claim_id values pass ingestion and are dropped later
files_reviewed:
  - actions/setup/js/aw_context.cjs
  - actions/setup/js/collect_ndjson_output.cjs
  - actions/setup/js/create_prompt.cjs
  - actions/setup/js/finish_work_queue_claim.cjs
  - actions/setup/js/generate_safe_outputs_tools.cjs
  - actions/setup/js/work_queue_mcp_server.cjs
  - actions/setup/js/work_queue_summary.cjs
  - actions/setup/js/write_work_queue_snapshot.cjs
  - pkg/workflow/compiler_safe_outputs_job.go
comment_count: 2

Note

This comment is managed by comment memory.

Expand the saved memory block to view or edit the persistent context for this thread.
Edit only the text inside the backtick fences; workflow metadata and the footer are regenerated automatically.

Learn more about comment memory

🔎 Code quality review by PR Code Quality Reviewer · copilot · gpt54 · 81 AIC · ⌖ 7.3 AIC · ⊞ 19.4K · ◷
_{Comment /review to run again}

github-actions

Skills-Based Review 🧠

Reviewed with /codebase-design and /tdd. Note: the pre-fetched diff (commit 128ee74) is one commit behind the current HEAD (f9a913a, "Reconcile downstream queue effects and fix compilation checks"). I verified the current checkout state directly instead.

Finding: both prior Copilot review comments on this PR are already resolved in f9a913a:

Claim-filtered agent_output.json is now consumed correctly by downstream jobs — upload_assets and safe-outputs.jobs now depend on safe_outputs and download the reconciled artifact (workQueueReconciledArtifactName) instead of the raw agent artifact, with ContinueOnError disabled for the fallback (pkg/workflow/safe_jobs.go, pkg/workflow/publish_assets.go, pkg/workflow/safe_outputs_steps.go).
TestWorkQueueCompilationPhases now asserts all_authorized for the user-step block and adds coverage for upload_assets/deploy gating on work_queue_output_types/work_queue_all_authorized.

I additionally confirmed reconcileWorkerClaim now derives a per-claim attempt identity (`${attempt}/claim/${encodeURIComponent(item.claim_id)}`) for multi-claim batches, which avoids Completion-transaction collisions when several claims are reconciled in the same workflow attempt — a correctness detail worth calling out as a strength.

📋 Verification performed

go build ./... — passes
go test ./pkg/workflow/... -run TestGateSafeOutputStepsForPartialWorkQueueClaims — passes
go test -tags integration ./pkg/workflow/... -run TestWorkQueueCompilationPhases — passes
go vet ./pkg/workflow/... — clean
JS vitest suites could not run locally (no network access for vitest install), but manual inspection of finish_work_queue_claim.test.cjs, work_queue_mcp_server.test.cjs, and write_work_queue_snapshot.test.cjs shows solid coverage of the new batch/array-claim paths (duplicate claim rejection, conflicting finish intents, per-claim filtering of agent_output.json, and claim_id schema injection into safe-output tool schemas).

✅ Positive highlights

Backward compatibility is preserved cleanly: single-claim behavior is unchanged (multiple branch is opt-in based on Array.isArray).
The "all-or-nothing" semantics for an omitted finish intent (cancelling the entire batch) is a sound, conservative security default and is clearly documented in work-queue.md and in-code comments.
Good symmetry between validWorker duplication in finish_work_queue_claim.cjs and work_queue_mcp_server.cjs — consider extracting to a shared helper in a follow-up to avoid drift, but not blocking.

No blocking issues found in the current HEAD state.

🧠 Reviewed using Matt Pocock's skills by Matt Pocock Skills Reviewer · copilot · sonnet50 · 82.7 AIC · ⌖ 14.4 AIC · ⊞ 10K
_{Comment /matt to run again}

github-actions · 2026-10-05T05:17:01Z

🏗️ Design Decision Gate — ADR Required

This PR adds 158 lines across business-logic directories (threshold: 100) and no existing ADR covers it, so ADR enforcement applies. I generated a draft ADR and pushed it to this branch.

Draft ADR: docs/adr/65682-claim-scoped-safe-outputs-for-batch-work-queue-claims.md (status: Draft)

Decision captured: accept an array of work-queue assignments in aw_context.work_queue and make safe outputs claim-scoped — outputs tagged with claim_id apply only once that claim's completion is verified; untagged outputs and custom steps/jobs require all claims to complete; a missing finish intent cancels the whole batch.

Evidence used

actions/setup/js/aw_context.cjs — readWorkQueueAssignment now accepts an array, rejecting duplicate claim_id/work_id.
actions/setup/js/finish_work_queue_claim.cjs (+96/−28) — per-claim reconciliation, claim-scoped attempt identity <attempt>/claim/<claim_id>, new partial status, filterClaimOutputs rewrite of agent_output.json, new all_authorized / output_types outputs.
actions/setup/js/work_queue_mcp_server.cjs — work_queue_claim_finish requires claim_id for batches and validates it against the trusted snapshot.
actions/setup/js/generate_safe_outputs_tools.cjs / collect_ndjson_output.cjs — optional claim_id on every safe-output tool, validated only when a work-queue snapshot exists.
.github/workflows/*.lock.yml — smoke test gate switched from authorized to all_authorized.

⚠️ Conflict with an existing ADR

ADR-64955 (Git-Backed Work Queue Coordination) states in its Worker protocol commitment:

Trusted aw_context.work_queue supplies one immutable inbound Claim ... each worker records at most one Completion and has one pass through safe-output processing.

This PR changes both commitments. ADR-64955 should be amended or superseded so the two records do not disagree.

Next action: review the drafted ADR, correct anything I inferred wrongly (especially the [TODO: verify] note about amending ADR-64955), then change its status from Draft to Proposed or Accepted.

🏗️ ADR gate enforced by Design Decision Gate 🏗️ · pi · opus50 · 84.1 AIC · ⌖ 50.4 AIC · ⊞ 6.7K · ◷
_{Comment /review to run again}

pelikhan · 2026-10-05T14:20:00Z

@copilot run pr-finisher skill

Copilot AI and others added 2 commits October 4, 2026 23:24

Support multiple queue claims and claim-tagged safe outputs

1e87165

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>

Cover partial-claim gating and single-item claim arrays

128ee74

Co-authored-by: pelikhan <4175913+pelikhan@users.noreply.github.com>

Copilot AI assigned Copilot and pelikhan Oct 4, 2026

Copilot AI changed the title ~~Support multi-claim work-queue workers and claim-scoped safe outputs~~ Support multiple work-queue claims and claim-scoped safe outputs Oct 4, 2026

Copilot finished work on behalf of pelikhan October 4, 2026 23:39

Copilot created this pull request from a session on behalf of pelikhan October 4, 2026 23:39 View session

Copilot AI requested a review from pelikhan October 4, 2026 23:39

github-actions Bot mentioned this pull request Oct 5, 2026

[PR Triage Report] PR triage report: work-queue and repo-memory reliability changes #65725

Closed

pelikhan marked this pull request as ready for review October 5, 2026 04:04

Copilot AI balanced review requested due to automatic review settings October 5, 2026 04:04

Copilot started reviewing on behalf of pelikhan October 5, 2026 04:04 View session

Copilot AI reviewed Oct 5, 2026

View reviewed changes

github-actions Bot mentioned this pull request Oct 5, 2026

[pr-sous-chef] Sous Chef Report: 1 PR nudged #65759

Closed

Copilot started work on behalf of gh-aw-bot October 5, 2026 04:40 View session

Copilot AI and others added 2 commits October 5, 2026 04:48

Merge remote-tracking branch 'origin/main' into copilot/update-work-q…

ded4e71

…ueue-api-support-multiple-claims Co-authored-by: gh-aw-bot <259018956+gh-aw-bot@users.noreply.github.com>

Reconcile downstream queue effects and fix compilation checks

f9a913a

Co-authored-by: gh-aw-bot <259018956+gh-aw-bot@users.noreply.github.com>

Copilot finished work on behalf of gh-aw-bot October 5, 2026 05:11

Copilot AI requested a review from gh-aw-bot October 5, 2026 05:11

github-actions Bot mentioned this pull request Oct 5, 2026

[aw] PR Code Quality Reviewer reported incomplete result #65780

Closed

github-actions Bot approved these changes Oct 5, 2026

View reviewed changes

docs(adr): add draft ADR-65682 for claim-scoped safe outputs

a5bc44b

This was referenced Oct 5, 2026

[PR Triage Report] PR triage report — 2026-10-05 #65799

Closed

[PR Triage Report] PR triage report - 2026-10-05 #65854

Closed

Copilot started work on behalf of pelikhan October 5, 2026 14:20 View session

pelikhan closed this Oct 5, 2026

Copilot stopped work on behalf of pelikhan due to an error October 5, 2026 14:34
The session was cancelled by the user.

Conversation

Copilot AI commented Oct 4, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI left a comment

Choose a reason for hiding this comment

Copilot review overview

🟡 Changes recommended

Uh oh!

Copilot AI Oct 5, 2026

Choose a reason for hiding this comment

Uh oh!

Copilot AI Oct 5, 2026

Choose a reason for hiding this comment

Uh oh!

gh-aw-bot commented Oct 5, 2026

Uh oh!

github-actions Bot commented Oct 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Oct 5, 2026

Uh oh!

github-actions Bot commented Oct 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Oct 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

github-actions Bot commented Oct 5, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Copilot AI commented Oct 5, 2026

Uh oh!

github-actions Bot commented Oct 5, 2026

Comment Memory

Uh oh!

github-actions Bot left a comment

Choose a reason for hiding this comment

Skills-Based Review 🧠

Uh oh!

github-actions Bot commented Oct 5, 2026

🏗️ Design Decision Gate — ADR Required

Uh oh!

pelikhan commented Oct 5, 2026

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

4 participants

Copilot AI commented Oct 4, 2026 •

edited

Loading

github-actions Bot commented Oct 5, 2026 •

edited

Loading

github-actions Bot commented Oct 5, 2026 •

edited

Loading

github-actions Bot commented Oct 5, 2026 •

edited

Loading

github-actions Bot commented Oct 5, 2026 •

edited

Loading