Codex review: needs maintainer review before merge. Reviewed May 31, 2026, 2:31 PM ET / 18:31 UTC.

Summary
The PR makes required auto-reply preflight compaction pass a forced preflight signal into embedded/context-engine compaction and adds user-facing recovery copy plus focused regression tests.

PR surface: Source +82, Tests +142. Total +224 across 9 files.

Reproducibility: yes. The current-main source path shows required preflight compaction can call the compactor without force, and the PR supplies production Feishu logs for the below-threshold preflight failure plus after-fix completion evidence.

Review metrics: 1 noteworthy metric.

• Compaction force surface: 1 existing public force flag reused; 3 internal preflight metadata fields added. Maintainers should notice that automatic preflight compaction now reaches the context-engine force contract, even though the reason/trigger metadata stays internal.

Merge readiness
Overall: 🐚 platinum hermit
Proof: 🦞 diamond lobster
Patch quality: 🐚 platinum hermit
Result: ready for maintainer review.

Overall follows the weaker of proof and patch quality, so missing proof can cap an otherwise strong patch.

Rank-up moves:

• Rebase or refresh the branch against current main and rerun the focused agent/auto-reply validation on the exact head.
• Get maintainer sign-off that forced automatic preflight compaction is the intended context-engine contract.

Risk before merge

• [P1] Forcing required preflight compaction changes the behavior from allowing a below-threshold context-engine skip to attempting compaction before prompt submission, which can rotate, checkpoint, or rewrite session state before the model call.
• [P1] Context-engine plugins that currently interpret force=true as manual or threshold compaction will now also see it for automatic preflight budget compaction, so maintainers need to accept that compatibility contract.
• [P1] The PR shares the auto-reply memory function with other open work and the live context did not show a clean mergeable head, so the exact head should be rebased/refreshed before landing.

Maintainer options:

1. Land after compaction owner sign-off (recommended)
   Accept that the preflight budget gate may force context-engine compaction before prompt submission, then rebase and rerun focused preflight/session tests on the exact head.
2. Narrow the force contract before merge
   If automatic force semantics are too broad, change the PR to force only engines that explicitly support preflight-required compaction or add a clearer internal mode.
3. Pause behind overlapping auto-reply work
   If the related auto-reply memory PRs land first, pause this branch until the shared function conflict is resolved and the diff can be reviewed fresh.

Next step before merge

• [P2] Human maintainer review is the right next step because the remaining blocker is accepting compatibility/session-state semantics and refreshing the merge head, not a narrow automated repair.

Security
Cleared: The diff does not change dependencies, workflows, secrets handling, package metadata, or external code execution surfaces; I found no concrete security or supply-chain concern.

ClawSweeper PR egg: 🎁 locked until real behavior proof passes.

Follow-up after ClawSweeper review:

• Addressed the ContextEngine public contract concern in 5cf430b20602eeb0a4695ec408879c4c2d78c8ed.
  • forceReason and preflightCompactionTrigger are no longer added to the public ContextEngine.compact() params type.
  • The public compact API remains additive only on the existing force?: boolean path.
  • The diagnostic reason/trigger now travel through internal runtimeContext, which avoids expanding the plugin SDK contract.
• Re-ran the focused regression suite locally after the change:

./node_modules/.bin/vitest run src/auto-reply/reply/agent-runner-memory.test.ts src/agents/embedded-agent-runner/compact.hooks.test.ts --maxWorkers=1

Test Files  3 passed (3)
Tests       144 passed (144)

Remaining known blocker:

• I have not added a Real behavior proof section yet because I do not want to overclaim. The branch has not been deployed into the live OpenClaw runtime, so I do not currently have honest after-fix evidence from a real oversized OpenClaw turn. The linked issue contains pre-fix live evidence (Preflight compaction required but failed: below threshold), while this PR currently has source-level regression proof. A real proof run will require either deploying this branch to a real setup and replaying/triggering an oversized preflight turn, or a maintainer applying proof: override if source+regression proof is sufficient for this narrow fix.

@clawsweeper re-review

🦞🧹
ClawSweeper re-review requested.

I asked ClawSweeper to review this item again.
Action: item re-review queued (workflow sweep.yml, event repository_dispatch).
Result: the existing ClawSweeper review comment will be edited in place when the review finishes.

Re-review progress:

• State: Complete
• Detail: The targeted re-review finished, the durable review comment was updated, and the synced verdict was routed.
• Run: https://github.com/openclaw/clawsweeper/actions/runs/26711481652
• Updated: 2026-05-31T11:40:26.992Z

Correction to my earlier proof update: a stronger live validation in the affected Feishu group just reproduced the original failure, so this should not be treated as proven yet.

Live production failure:

chat: oc_5cc1621461880ae8dccec0147e40896b (Merlin 进化论)
session: agent:main:feishu:group:oc_5cc1621461880ae8dccec0147e40896b
message: om_x100b6e49341518b4c398dd246766af6

2026-05-28T13:50:41.213+08:00 message dispatch completed: channel=feishu sessionKey=agent:main:feishu:group:oc_5cc1621461880ae8dccec0147e40896b outcome=error error="Error: Preflight compaction required but failed: below threshold"
2026-05-28T13:50:41.217+08:00 message processed: channel=feishu chatId=chat:oc_5cc1621461880ae8dccec0147e40896b messageId=om_x100b6e49341518b4c398dd246766af6 outcome=error error="Error: Preflight compaction required but failed: below threshold"
2026-05-28T13:50:41.222+08:00 feishu[default][msg:om_x100b6e49341518b4c398dd246766af6]: dispatch failed: Error: Preflight compaction required but failed: below threshold

The earlier simple probe turns (FEISHU_DM_OK / FEISHU_EVOLUTION_OK) were not representative enough; they proved simple dispatch still worked, but not the oversized preflight-vs-threshold path. I updated the PR body accordingly and will treat this as still needing a follow-up fix before merge.

Correction/update to the earlier failed-proof note:

The failed live validation at 13:50 was caused by a stale running gateway process, not by the installed package contents after restart.

• Installed package / CLI: 2026.5.26
• Stale gateway during the failure: 2026.5.22, PID 3752207, started Wed May 27 09:52:55 2026
• After real systemd restart: Gateway 2026.5.26, PID 3953195, started Thu May 28 14:55:53 2026, connectivity probe ok

Pre-restart failure in the affected Feishu group:

2026-05-28T13:50:41.213+08:00 message dispatch completed: channel=feishu sessionKey=agent:main:feishu:group:oc_5cc1621461880ae8dccec0147e40896b outcome=error error="Error: Preflight compaction required but failed: below threshold"

Post-restart validation in the same affected Feishu group:

2026-05-28T15:12:30.587+08:00 feishu[default]: dispatching to agent (session=agent:main:feishu:group:oc_5cc1621461880ae8dccec0147e40896b)
2026-05-28T15:16:58.070+08:00 feishu[default][msg:om_x100b6e4ae31ba894c32594bf37c25d4]: phase transition (from=streaming, to=completed, source=onIdle, reason=normal)
2026-05-28T15:16:59.089+08:00 feishu[default][msg:om_x100b6e4ae31ba894c32594bf37c25d4]: reply completed, card finalized (elapsedMs=268502, isCardKit=true)

A subsequent real project turn (merge 483，下一步) also produced a visible Feishu CardKit reply and continued the workflow. No post-restart occurrence of Preflight compaction required but failed: below threshold was observed for the affected group.

I updated the PR body accordingly. Scope boundary: this is proof for the below threshold preflight failure path after loading the fixed runtime; it is not a blanket claim that every future high-context overflow case is solved.

@clawsweeper re-review

🦞👀
ClawSweeper picked this up.

Command router queued. I will update this comment with the next step.

Re-review progress:

• State: Complete
• Detail: The targeted re-review finished, the durable review comment was updated, and the synced verdict was routed.
• Run: https://github.com/openclaw/clawsweeper/actions/runs/26560863651
• Updated: 2026-05-28T07:32:29.382Z

Fresh related observation from the same production incident window:

• The original Preflight compaction required but failed: below threshold path remains the key root-cause class for oversized Feishu sessions (Feishu DM messages visible in Feishu history but missing from OpenClaw session #87234).
• In addition, Feishu text control commands (/compact / compact, and new) also showed user-visible failures/no-response in the same timeframe. Those are tracked separately on Bug: Feishu channel does not handle text slash commands (/compact, /new, /stop) #79409 / fix(auto-reply): fast path text control commands #85616 and should not be collapsed into this PR.

For this PR specifically, I agree the remaining blocker is exact-head/head-built proof for the oversized preflight path. The production restart proof showed the installed package/runtime behavior after restart, but the maintainer concern that it was not exact PR-head proof is fair.

Status refresh after checking current PR checks without pulling the repo locally:

• The latest Real behavior proof check is now passing.
• dependency-guard is currently skipped on the latest visible run, not a current hard failure.
• The remaining hard failing check I can see is checks-node-agentic-control-plane-runtime.
• Its log points at two gateway shared-token suites failing during module mocking/hoisting, not at the preflight compaction patch itself:
  • src/gateway/server.shared-token-hot-reload.test.ts
  • src/gateway/server.shared-token-session-rotation.test.ts
  • error: [vitest] There was an error when mocking a module... make sure there are no top level variables inside, since this call is hoisted

Given the proof check is now green, this may need either a maintainer rerun/refresh of the unrelated runtime shard or a separate fix for those gateway shared-token tests. I’m not treating this as exact-head merge-ready yet because the status label still says needs proof and the runtime shard is red.

@clawsweeper re-review

🦞👀
ClawSweeper picked this up.

Command router queued. I will update this comment with the next step.

Land-ready after maintainer fixups.

Summary:

• Forced preflight compaction now treats required compaction as a hard gate: compact, skip only explicit harmless no-op reasons, or fail with Preflight compaction required but failed: ....
• Direct and queued reply paths now surface that preflight-required failure as a visible source-suppression-safe recovery message before starting the agent.
• Unrelated queued preflight failures still reject instead of being converted into a compaction notice.

Local proof:

• node scripts/run-vitest.mjs src/auto-reply/reply/agent-runner-direct-runtime-config.test.ts src/auto-reply/reply/followup-runner.test.ts src/auto-reply/reply/agent-runner-memory.test.ts src/agents/embedded-agent-runner/compact.hooks.test.ts --maxWorkers=1 passed: 2 Vitest shards, 180 tests.
• .agents/skills/autoreview/scripts/autoreview --mode branch --base origin/main --prompt ... passed clean: no accepted/actionable findings.

CI proof:

• CI run 26721201572 passed, including preflight, check-guards, check-lint, check-test-types, checks-node-agentic-agents-core, checks-node-auto-reply-reply-agent-runner, and ci-timings-summary.
• Real behavior proof run 26721206824 passed.
• Critical Quality agent-runtime-boundary run 26721201571 passed.
• CodeQL Security High jobs, OpenGrep PR diff, dependency guard, workflow sanity, labeler, and auto-response passed.

Known proof gaps: none.