fix(agents): preserve sibling reroutes and repair Gateway release E2E

[steipete]

What Problem This Solves

Resolves a problem where the 2026.9.1 stable release's Gateway E2E validation failed, including a requester wake that dropped a sibling's model-route change notice.

Failure evidence: Full Release Validation, Gateway 4/4 job, release SHA c92d0ae9c07e757719e676517aa9f9460a40761c.

Why This Change Was Made

Requester settle wakes now collect route notices from the same current, deduplicated children as completion findings. Notices are sorted, deduplicated, and bounded to 1,024 characters including an explicit truncation marker. The existing local/shared-channel visibility policy remains authoritative.

The other failures came from fixtures that no longer described current owner contracts:

• The App SDK worker-service fixture hid missing required methods behind a context cast. It now satisfies WorkerEnvironmentServiceContract and proves execution-mode projection. The call was introduced by fix: prevent incompatible cloud runtime dispatches #126585; fix: keep cloud sessions stopped and surface cleanup failures #135583 removed the method-level optional guard. The real service already implements it.
• Requester and sibling lifecycle fixtures returned assistant text without a confirmed outbound delivery receipt. They now return terminal send receipts and wait for committed delivery state; failed assertions also release gated delivery work.
• Doctor's reasonless skipped update is a failed update, not a continuation. The fixture now explicitly supplies the supported already-current no-op and retains the schema-refusal assertion.
• Four sibling lifecycle cases dated their fresh events in 1970 while the fake clock was in 2026, immediately hitting the delivery deadline. Event timestamps now come from the fixture clock; grace timing and assertions are preserved.

Each independent cause has its own commit. The change restores the existing subagent handoff and delivery contracts without adding configuration, persistence, schema, or protocol changes. Production delta is +27/-10; the net 17 lines implement current-batch aggregation, deterministic ordering, and a bounded prompt block.

User Impact

Parents receive route-change notices from rerouted siblings even when another child triggers the consolidated wake. External-channel parents continue to keep those notices internal. Release validation now exercises realistic environment, update, and delivery contracts.

Related: #136390 addresses overlapping-wave deferral, a separate invariant; this PR does not include that change.

Evidence

All four original failures reproduced in 3/3 runs before edits and passed 3/3 after repairs, using test/vitest/vitest.e2e.config.ts, two workers, and the prepared E2E runtime. Fixture-only proof fixed App SDK, doctor, and rejected-wake tests while the visible-wake test still failed on the missing sibling notice before the production fix.

The final combined E2E run passed 22/22 tests across four files, including the sibling lifecycle suite. The eight initial sibling failures reduced to four after receipt fixes, then zero after correcting event timestamps.

OPENCLAW_E2E_WORKERS=2 OPENCLAW_E2E_USE_PREBUILT_DIST=1 \
node scripts/run-vitest.mjs run --config test/vitest/vitest.e2e.config.ts \
  packages/sdk/src/app-sdk-composed-resources.e2e.test.ts \
  src/agents/subagents/registry/subagent-registry.requester-wake.e2e.test.ts \
  src/commands/doctor.refuses-newer-database-schemas.e2e.test.ts \
  src/agents/subagents/registry/subagent-registry.lifecycle-retry-grace.e2e.test.ts

The initial invocation omitted OPENCLAW_E2E_USE_PREBUILT_DIST and successfully built the runtime, local plugin artifacts, and subprocess prerequisites. Local proof used macOS and Node 26.8.1; the release job uses Linux and Node 24.19.0.

A recent scheduled main E2E run already had the exact App SDK failure in shard 1 and doctor failure in shard 2. Weighted assignment differed; the requester-wake file was absent from that earlier run. Normal ci.yml does not run this full Gateway E2E matrix.

Owner unit suites passed 107/107 (39 requester-wake, 33 environment, 35 doctor-update). The final requester-wake suite passed again after test consolidation; the two deferred delivery cases also passed after adopting the existing promise helper.

node scripts/run-vitest.mjs \
  src/agents/subagents/announce/subagent-announce.requester-settle-wake.test.ts \
  src/gateway/server-methods/environments.test.ts \
  src/commands/doctor-update.test.ts
pnpm check:changed

pnpm check:changed completed successfully with every selected gate enabled. Earlier attempts identified test-file line limits and an unsupported TypeScript library reference in the test consolidation; duplicate setup was consolidated and the canonical deferred-promise helper reused. No assertions, gates, or type targets were weakened. Final independent Codex review found no actionable P0–P2 issues. Production is +27/-10; tests are +206/-192. The complete Gateway shard and full release matrix have not been rerun; this PR is the focused repair and local proof for the release owner to integrate.

🤖 Generated with Claude Code

[clawsweeper]

🦞👀
ClawSweeper picked this up.

Pull request received. I will update this pull request when review starts.

[clawsweeper]

Codex review: blocked before merge. Reviewed September 2, 2026, 11:49 PM ET / September 3, 2026, 03:49 UTC.

ClawSweeper review

What this changes

The branch collects and bounds route-change notices from all current sibling subagent completions in a consolidated requester wake, while updating Gateway E2E fixtures to current contracts.

Merge readiness

⛔ Blocked before merge - 2 items remain

Keep this actively maintained, maintainer-labeled PR open. Source review found no introduced blocking defect: it repairs the current-main scheduling-row-only route-notice loss and aligns E2E fixtures with current delivery, environment, and doctor contracts.

Priority: P2
Reviewed head: e56f922ae132f19ac2be9662095eb0df1e392aaf
Owner decision: Required. See Decision needed.

Review scores

Measure	Result	What it means
Overall readiness	🐚 platinum hermit (4/6)	A focused owner-boundary repair with aligned fixture updates and no blocking source finding.
Proof confidence	🌊 off-meta tidepool	Not applicable: The repository state identifies this as a MEMBER-authored PR, so the external-contributor proof gate does not apply; the supplied body nevertheless records exact-head focused Gateway E2E before/after runs, which were not re-executed in this read-only review.
Patch quality	🐚 platinum hermit (4/6)	No actionable review findings were identified.

Verification

Check	Result	Evidence
Real behavior	Not applicable	Not applicable: The repository state identifies this as a MEMBER-authored PR, so the external-contributor proof gate does not apply; the supplied body nevertheless records exact-head focused Gateway E2E before/after runs, which were not re-executed in this read-only review.
Evidence reviewed	7 items	Current-main defect: Current main builds completion findings from filtered, deduplicated rows but reads the route-change notice only from the scheduling child, so a different rerouted sibling can be omitted. Introduced repair: The PR reuses the same current-child selection for findings and route notices, then deduplicates, sorts, and caps notices at 1,024 UTF-16 code units. Lifecycle caller: The lifecycle scheduler supplies the settled entry and requester origin to the wake owner, establishing why the scheduling child can differ from the sibling that owns the route notice.
Findings	None	None.
Security	None	None.

How this fits together

Subagent completion records flow from the registry into a consolidated wake for the parent agent. That wake builds model-visible completion context and returns it through delivery policy, which decides whether route notices remain internal on shared channels.

flowchart LR
  A[Child subagent completions] --> B[Subagent registry]
  B --> C[Current child selection]
  C --> D[Consolidated requester wake]
  D --> E[Channel visibility policy]
  E --> F[Parent agent context]

Loading

Decision needed

Question	Recommendation
Should this maintainer-owned repair be landed after the exact-head required checks complete successfully?	Land after green exact-head checks: Accept the focused repair because current main still has the scheduling-row-only behavior and the branch covers the sibling route-notice path.

Why: The PR is explicitly maintainer-labeled and the repository policy reserves protected maintainer work for explicit maintainer handling, even though source review found no blocking patch defect.

Before merge

• Complete next step (P2) - A maintainer should decide whether to land the exact reviewed head after its required exact-head checks finish green.
• Resolve maintainer decision - Resolve the maintainer decision shown above before merge.

Agent review details

Security

None.

PR surface

Source +17, Tests +14. Total +31 across 6 files.

View PR surface stats

Area	Files	Added	Removed	Net
Source	1	27	10	+17
Tests	5	206	192	+14
Docs	0	0	0	0
Config	0	0	0	0
Generated	0	0	0	0
Other	0	0	0	0
Total	6	233	202	+31

Review metrics

Metric	Value	Why it matters
Production versus test delta	production +17 net; tests +14 net	The production addition is confined to deterministic, bounded aggregation from an existing selected batch; the remaining change refreshes focused E2E and unit fixtures.

Technical review

Best possible solution:

Land a single canonical wake formatter that derives completion findings and route notices from the same current-child batch, retaining the bounded deterministic aggregation and boundary-level coverage.

Do we have a high-confidence way to reproduce the issue?

Yes, source-reproducible: construct a settled batch where the scheduling child has no route notice and a current sibling does; fetched main reads only the scheduling child at the wake formatter.

Is this the best way to solve the issue?

Yes: deriving both findings and notices from the existing filtered, deduplicated child selection keeps one authoritative batch boundary and avoids a sibling-specific parallel path.

AGENTS.md: found and applied where relevant.

Codex review notes: model internal, reasoning high; reviewed against 4af3fc294575.

Labels

Label justifications:

• P2: This is a bounded repair to parent-agent completion context and the associated Gateway E2E coverage.
• rating: 🐚 platinum hermit: Overall readiness is 🐚 platinum hermit; proof is 🌊 off-meta tidepool and patch quality is 🐚 platinum hermit.
• status: 👀 ready for maintainer look: ClawSweeper has no concrete contributor-facing blocker left for this PR. Not applicable: The repository state identifies this as a MEMBER-authored PR, so the external-contributor proof gate does not apply; the supplied body nevertheless records exact-head focused Gateway E2E before/after runs, which were not re-executed in this read-only review.

Evidence

What I checked:

• Current-main defect: Current main builds completion findings from filtered, deduplicated rows but reads the route-change notice only from the scheduling child, so a different rerouted sibling can be omitted. (src/agents/subagents/announce/subagent-announce.requester-settle-wake.ts:398, 4af3fc294575)
• Introduced repair: The PR reuses the same current-child selection for findings and route notices, then deduplicates, sorts, and caps notices at 1,024 UTF-16 code units. (src/agents/subagents/announce/subagent-announce.requester-settle-wake.ts:390, e56f922ae132)
• Lifecycle caller: The lifecycle scheduler supplies the settled entry and requester origin to the wake owner, establishing why the scheduling child can differ from the sibling that owns the route notice. (src/agents/subagents/registry/subagent-registry-lifecycle-wake.ts:323, e56f922ae132)
• Boundary coverage: The product-flow E2E creates two visible children, gives the second child a route notice, and asserts that the consolidated shared-surface wake retains the notice while keeping it internal. (src/agents/subagents/registry/subagent-registry.requester-wake.e2e.test.ts:319, e56f922ae132)
• Fixture-contract alignment: The SDK fixture now satisfies the complete request-facing worker-environment contract and asserts execution-mode projection; the production handler requires that method when listing profiles. (packages/sdk/src/app-sdk-composed-resources.e2e.test.ts:135, e56f922ae132)
• Codex boundary check: Direct Codex inspection shows its EnvironmentManager owns Codex execution/filesystem environments; this PR does not call or alter that protocol, and its environment change is confined to an OpenClaw test fixture. (codex-rs/exec-server/src/environment.rs:69, 5adb68a49933)

Likely related people:

• steipete: Suggested for follow-up; no historical authorship or introduction is verified. (role: unverified routing candidate; confidence: low)
• Mert Başar: Suggested for follow-up; no historical authorship or introduction is verified. (role: unverified routing candidate; confidence: low)

Rating scale

Score	Internal tier	Crab rank	Meaning
6/6	S	🦀 challenger crab	Exceptional readiness
5/6	A	🦞 diamond lobster	Very strong readiness
4/6	B	🐚 platinum hermit	Good normal PR; ordinary maintainer review
3/6	C	🦐 gold shrimp	Useful, but confidence is limited
2/6	D	🦪 silver shellfish	Proof or implementation needs work
1/6	F	🧂 unranked krab	Not merge-ready
N/A	NA	🌊 off-meta tidepool	Rating does not apply

Overall follows the weaker of proof and patch quality.
Shiny media proof means a screenshot, video, or linked artifact directly shows the changed behavior. Runtime, network, CSP, and security claims still need visible diagnostics.

Workflow

• ClawSweeper keeps one durable marker-backed review comment per issue or PR.
• Re-runs edit this comment so the latest verdict, findings, and automation markers stay together instead of adding duplicate bot comments.
• A fresh review can be triggered by eligible @clawsweeper re-review comments, exact-item GitHub events, scheduled/background review runs, or manual workflow dispatch.
• PR/issue authors and users with repository write access can comment @clawsweeper re-review or @clawsweeper re-run on an open PR or issue to request a fresh review only.
• Maintainers can also comment @clawsweeper review to request a fresh review only.
• Fresh-review commands do not start repair, autofix, rebase, CI repair, or automerge.
• Maintainer-only repair and merge flows require explicit commands such as @clawsweeper autofix, @clawsweeper automerge, @clawsweeper fix ci, or @clawsweeper address review.
• Maintainers can comment @clawsweeper explain to ask for more context, or @clawsweeper stop to stop active automation.

History

Review history (2 earlier review cycles)

• reviewed 2026-09-03T01:43:51.966Z sha cab7419 :: blocked before merge. :: none
• reviewed 2026-09-03T03:08:00.258Z sha 7a3f5a6 :: blocked before merge. :: none

[steipete]

Merged via squash.

• Prepared head SHA: e56f922ae132f19ac2be9662095eb0df1e392aaf
• Landed commit: 5ecfe38102ce72e723b743b0f2b2a2ef45d479d5