Best possible solution:

Rebase onto current main, drop unrelated history, guard only timeout-triggered compaction with externalAbort, preserve timeout error payload synthesis, add focused regression coverage, and attach redacted real Control UI Stop proof.

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

Yes by source inspection: current main can enter timeout compaction on a timed-out user abort because the compaction gate ignores externalAbort. The PR-introduced payload regression is also source-reproducible from the head file contents and existing payload helpers.

Is this the best way to solve the issue?

No. externalAbort is the right discriminator for the compaction gate, but applying it to timedOutDuringPrompt suppresses the timeout error payload, and the branch must be narrowed before it is reviewable.

Full review comments:

• [P1] Rebase away stale branch-wide reversions — CHANGELOG.md:4
  The PR head is not a narrow timeout fix: comparing it to current main replaces the 2026.5.14 changelog with a small Unreleased section, and the API reports 5,205 changed files. Merging this branch as-is would drop current release notes and broad current-main changes; rebase onto current main and keep only the focused runner/test diff.
  Confidence: 0.96
• [P2] Preserve timeout payload synthesis — src/agents/pi-embedded-runner/run.ts:2376
  This guard makes timedOutDuringPrompt false for the new timedOut/aborted/externalAbort case. Since the payload builder has no timeout fallback and incomplete-turn synthesis returns null for aborted or timed-out attempts, the run can fall through without the timeout error payload the new test expects.
  Confidence: 0.93

Overall correctness: patch is incorrect
Overall confidence: 0.93

Security concerns:

• [high] Remove unrelated release and dependency churn — .github/workflows/openclaw-release-publish.yml:53
  The branch touches release workflows, CodeQL configuration, Docker/package surfaces, and many dependency files while the stated fix only needs agent runner/test changes. The current-main comparison also shows release workflow downgrades/removals, which is supply-chain sensitive if merged accidentally.
  Confidence: 0.91

What I checked:

• Live PR metadata: GitHub API reports the PR is open, not draft, mergeable=false, head 17b3165, with 5,205 changed files, 160,383 additions, 152,506 deletions, the protected security label, and triage: needs-real-behavior-proof. (17b3165a974d)
• Focused compaction change: The PR head adds && !externalAbort to the timeout-triggered compaction condition, which is the right discriminator for user-initiated stops versus internal provider timeouts. (src/agents/pi-embedded-runner/run.ts:1450, 17b3165a974d)
• Current main behavior: Current main still runs timeout-triggered compaction for timedOut && !timedOutDuringCompaction && !timedOutDuringToolExecution without checking externalAbort, so the central bug claim is source-reproducible. (src/agents/pi-embedded-runner/run.ts:1569, 99a6b1c5a886)
• Timeout payload regression: The PR also adds !externalAbort to timedOutDuringPrompt; with timedOut=true, aborted=true, externalAbort=true, and no assistant text, this bypasses the explicit timeout error payload path. (src/agents/pi-embedded-runner/run.ts:2376, 17b3165a974d)
• No fallback timeout payload: resolveIncompleteTurnPayloadText returns null when the attempt is aborted or timed out, and buildEmbeddedRunPayloads builds payloads from assistant/tool content rather than timeout state, so the guarded timeout path is required for the added test expectation. (src/agents/pi-embedded-runner/run/incomplete-turn.ts:237, 17b3165a974d)
• Stale branch comparison: Comparing the PR head to current main shows the branch would replace the current 2026.5.14 changelog with a small Unreleased section, confirming this is not a narrow mergeable diff against main. (CHANGELOG.md:4, 17b3165a974d)

Likely related people:

• vincentkoc: Available current-main blame points the timeout compaction gate, timeout payload synthesis, and abort flag plumbing at the current snapshot, and recent path history shows additional runner/attempt maintenance by this contributor. (role: recent area contributor; confidence: medium; commits: 278e3eabf2, a47132734ba2, af3d9333aa8a; files: src/agents/pi-embedded-runner/run.ts, src/agents/pi-embedded-runner/run/attempt.ts)
• steipete: Recent GitHub path history for timeout-compaction tests and timeout policy includes several focused commits in this area. (role: timeout-compaction test and policy contributor; confidence: medium; commits: 908bb0f1c74e, f3e8a8a3195a, d9dc75774bcb; files: src/agents/pi-embedded-runner/run.timeout-triggered-compaction.test.ts, src/agents/pi-embedded-runner/run.ts)
• doedewaldt: Recent run-loop history includes explicit timeout error and partial-fragment handling commits that are directly related to the PR's timeout payload path. (role: recent timeout payload contributor; confidence: medium; commits: db57da50c994, 18f4a2d98723; files: src/agents/pi-embedded-runner/run.ts)

Remaining risk / open question:

• No after-fix Control UI Stop proof is present, so the claimed user-facing behavior has not been demonstrated in a real setup.
• The branch is far wider than the described fix and includes security-sensitive workflow and dependency surfaces that need pruning or separate review.

Codex review notes: model gpt-5.5, reasoning high; reviewed against 99a6b1c5a886.