• 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.

• 🦀 challenger crab: rare, exceptional readiness with strong proof, clean implementation, and convincing validation.
• 🦞 diamond lobster: very strong readiness with only minor maintainer review expected.
• 🐚 platinum hermit: good normal PR, likely mergeable with ordinary maintainer review.
• 🦐 gold shrimp: useful signal, but proof or patch confidence is still limited.
• 🦪 silver shellfish: thin signal; proof, validation, or implementation needs work.
• 🧂 unranked krab: not merge-ready because proof is missing/unusable or there are serious correctness or safety concerns.
• 🌊 off-meta tidepool: rating does not apply to this item.

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.

Best possible solution:

Keep the current explicit delivery contract from #84397; if a prompt wording regression exists, replace it with narrower wording that still preserves the before-ending message(action="send") requirement and private-final semantics.

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

We have a high-confidence source-level reproduction of the patch regression: current tests and prompt snapshots require explicit message-tool delivery guidance, while the PR replaces it with generic wording. I did not run a live Telegram or Discord turn in this read-only review.

Is this the best way to solve the issue?

No. Reverting the current fix is not the best solution because it removes the explicit prompt contract that keeps message-tool-only Codex replies visible; a narrower, live-proven wording adjustment would be safer if a maintainer wants different copy.

Label changes:

• add P1: The PR would regress an agent/channel reply path where visible user-facing messages can remain private instead of being delivered.
• add merge-risk: 🚨 message-delivery: The diff removes the prompt contract that tells Codex how to visibly send replies when automatic source delivery is suppressed.
• add rating: 🧂 unranked krab: Current PR rating is 🧂 unranked krab because proof is 🧂 unranked krab, patch quality is 🧂 unranked krab, and The PR reverts a current message-delivery fix, has a blocking correctness finding, and lacks real behavior proof.
• add status: 📣 needs proof: The PR needs real behavior proof before ClawSweeper can clear the contributor ask. Needs real behavior proof before merge: No after-change real behavior proof is present; before merge, add redacted Telegram/Discord proof or live logs showing the message-tool-only visible reply path and update the PR body so ClawSweeper can re-review.
• add mantis: telegram-visible-proof: Mantis should capture Telegram visible proof. This touches Telegram message-tool-only prompt snapshots and visible reply behavior that a short Telegram Desktop proof can demonstrate.

Label justifications:

• P1: The PR would regress an agent/channel reply path where visible user-facing messages can remain private instead of being delivered.
• merge-risk: 🚨 message-delivery: The diff removes the prompt contract that tells Codex how to visibly send replies when automatic source delivery is suppressed.
• rating: 🧂 unranked krab: Current PR rating is 🧂 unranked krab because proof is 🧂 unranked krab, patch quality is 🧂 unranked krab, and The PR reverts a current message-delivery fix, has a blocking correctness finding, and lacks real behavior proof.
• status: 📣 needs proof: The PR needs real behavior proof before ClawSweeper can clear the contributor ask. Needs real behavior proof before merge: No after-change real behavior proof is present; before merge, add redacted Telegram/Discord proof or live logs showing the message-tool-only visible reply path and update the PR body so ClawSweeper can re-review.
• mantis: telegram-visible-proof: Mantis should capture Telegram visible proof. This touches Telegram message-tool-only prompt snapshots and visible reply behavior that a short Telegram Desktop proof can demonstrate.

Full review comments:

• [P1] Keep the explicit message-tool delivery instruction — extensions/codex/src/app-server/thread-lifecycle.ts:887
  This replaces the message-tool-only Codex guidance with a generic use the message tool sentence. In this mode automatic source delivery is suppressed and normal final text is private, so dropping the action="send" before-ending instruction reintroduces the hidden-reply path that the reverted fix and changelog tie to direct replies staying private.
  Confidence: 0.93

Overall correctness: patch is incorrect
Overall confidence: 0.91

What I checked:

• Current main Codex prompt guard: Current main's buildVisibleReplyInstruction tells message-tool-only Codex turns to preserve channel/session context, call message with action="send" before ending, not rely on private final text, and avoid the send call when no visible response is needed. (extensions/codex/src/app-server/thread-lifecycle.ts:887, 5c39e0019d2f)
• PR diff removes the explicit guard: The PR head replaces the four-part message-tool-only Codex instruction with the generic sentence To send a visible message, use the message tool, dropping the before-ending action call and private-final warning. (extensions/codex/src/app-server/thread-lifecycle.ts:887, 950e7d5b056b)
• Tool description still encodes the delivery contract: Current main's message tool description for message_tool_only includes action="send" before ending and says normal final answers stay private and are not visible in this mode. (src/agents/tools/message-tool.ts:862, 5c39e0019d2f)
• Tests protect the explicit wording: The current Codex app-server test asserts that developer instructions include the explicit message/action="send" before-ending text and the private-final warning for message-tool-only delivery. (extensions/codex/src/app-server/run-attempt.test.ts:1405, 5c39e0019d2f)
• Changelog ties the current behavior to the reported bug: The changelog says the Codex/message fix tells message-tool-only Codex turns to send visible output with message(action="send") before ending so direct replies do not stay private, and links [Bug]: v2026.5.18: Telegram plain-text outbound dispatch silently fails after memory-pressure event; survives docker restart; cron and slash-command paths still work #84129 plus fix: prompt Codex to send visible channel replies #84397. (CHANGELOG.md:32, 5c39e0019d2f)
• Fix provenance: Git history identifies 47eb4ca as the current-main fix commit for fix: prompt Codex to send visible channel replies #84397, committed on 2026-05-19T23:29:41-05:00. (47eb4ca14fcc)

Likely related people:

• Josh Avant: git blame shows the explicit message-tool-only delivery guidance in the Codex prompt, message tool description, inbound metadata, and group/direct prompt text came from the merged fix commit for fix: prompt Codex to send visible channel replies #84397. (role: introduced current fix; confidence: high; commits: 47eb4ca14fcc; files: extensions/codex/src/app-server/thread-lifecycle.ts, src/agents/tools/message-tool.ts, src/auto-reply/reply/inbound-meta.ts)
• Dallin Romney: The shallow current history shows the surrounding message-tool-only prompt and tool-description scaffolding in the grafted baseline commit, so this is a relevant adjacent history signal rather than sole ownership. (role: earlier area contributor; confidence: medium; commits: 3d96111a5a; files: extensions/codex/src/app-server/thread-lifecycle.ts, src/agents/tools/message-tool.ts, src/auto-reply/reply/groups.ts)
• pash-openai: git blame shows adjacent buildVisibleReplyInstruction structure in thread-lifecycle.ts came from the recent Codex developer-instruction move, so this person is relevant for Codex prompt-layer routing beyond the PR branch itself. (role: recent adjacent Codex prompt contributor; confidence: medium; commits: e0d1a2a9b9; files: extensions/codex/src/app-server/thread-lifecycle.ts)

Codex review notes: model gpt-5.5, reasoning high; reviewed against 5c39e0019d2f.

• The egg game starts only after the PR passes the real-behavior proof check.
• Before that, no creature or rarity is rolled. The treat waits for real proof.
• This is still just collectible flavor: proof affects review readiness, not creature quality.