Codex usage limits have been reached for code reviews. Please check with the admins of this repo to increase the limits by adding credits.
Repo admins can enable using credits for code reviews in their settings.

Greptile Summary

This PR fixes a bug where agent runs were emitting both intermediate tool-use assistant text and the final answer as user-visible replies. It expands shouldSuppressAssistantVisibleOutput to handle unphased messages by checking for stopReason: "toolUse" or tool-call content blocks, and updates the mid-stream second gate in handleMessageUpdate to use the same predicate. A regression test is added for the unphased tool-use path.

Confidence Score: 4/5

Safe to merge; the fix is well-scoped and the two P2 findings are non-blocking.

Only P2-level findings: a gap in test coverage for the text_end channel and a theoretical miss of snake_case content-type variants. The core suppression logic is correct and the existing commentary phase behaviour is preserved.

No files require special attention beyond the P2 notes above.

This is a comment left during a code review.
Path: src/agents/pi-embedded-subscribe.subscribe-embedded-pi-session.suppresses-commentary-phase-output.test.ts
Line: 10-40

Comment:
**Missing `blockReplyBreak: "text_end"` coverage**

The new regression test only exercises the `blockReplyBreak: "message_end"` path. The `handleMessageUpdate` code has a separate `text_end`-channel flush at lines 617–634 (`ctx.blockChunker?.drain` and the deferred `flushBlockReplyBuffer`). A parallel fixture with `blockReplyBreak: "text_end"` would confirm those paths are also suppressed for unphased tool-use messages and guard against future regressions there.

How can I resolve this? If you propose a fix, please make it concise.

---

This is a comment left during a code review.
Path: src/agents/pi-embedded-subscribe.handlers.messages.ts
Line: 40

Comment:
**Snake-case content-type variants not covered**

`TOOL_CALL_CONTENT_TYPES` contains `"toolCall"`, `"toolUse"`, and `"functionCall"` (camelCase). Some providers (e.g. OpenAI-compatible adapters) may emit `"tool_call"`, `"tool_use"`, or `"function_call"` (snake_case). If any upstream adapter normalises content block types to snake_case before emitting events, the content-block branch of `hasToolUseContinuation` would miss them. The `stopReason === "toolUse"` path would still fire, so this only matters when `stopReason` is absent and the detection falls back to content scanning.

How can I resolve this? If you propose a fix, please make it concise.

_{Reviews (1): Last reviewed commit: "fix(agent): suppress tool-use assistant ..." | Re-trigger Greptile}

Rebased this branch on current main to pick up unrelated CI fixes that landed upstream after the previous run (notably the Array#toSorted() lint fix).\n\nNew head: 83f7a5dc1a.\n\nLocal validation after rebase:\n- corepack pnpm vitest run src/agents/pi-embedded-subscribe.subscribe-embedded-pi-session.suppresses-commentary-phase-output.test.ts\n- corepack pnpm tsgo:core\n- corepack pnpm lint:core\n- corepack pnpm exec oxfmt --check src/agents/pi-embedded-subscribe.handlers.messages.ts src/agents/pi-embedded-subscribe.subscribe-embedded-pi-session.suppresses-commentary-phase-output.test.ts

Pushed a follow-up commit: e8f55349fe (fix(agent): clear suppressed commentary buffers).

This addresses the remaining gap behind #71588. The existing suppression path covers commentary/tool-use assistant text once the phase is already resolved, but it does not fully cover the delayed-phase stream case:

• text_delta can arrive before the stream exposes phase: "commentary" via text_end / textSignature.
• In that window, the visible assistant text may already have been appended into assistantTexts, blockBuffer, or partial/block chunking state.
• Returning early in handleMessageEnd is then too late: the user-visible buffers can already be contaminated, which can still produce an intermediate/tool-use-looking reply alongside the final answer.

The new commit keeps the narrow suppression behavior, but additionally clears suppressed delta/block/chunker buffers and rolls assistantTexts back to the pre-block baseline when a commentary/tool-use assistant block is suppressed. It also expands the tool-call block guard to include snake_case variants (tool_call, tool_use, function_call).

Local validation:

• corepack pnpm vitest run src/agents/pi-embedded-subscribe.subscribe-embedded-pi-session.suppresses-commentary-phase-output.test.ts
• corepack pnpm tsgo:core
• corepack pnpm lint:core
• corepack pnpm exec oxfmt --check src/agents/pi-embedded-subscribe.handlers.messages.ts src/agents/pi-embedded-subscribe.subscribe-embedded-pi-session.suppresses-commentary-phase-output.test.ts

Thanks for the context here. I swept through the related work, and this is now duplicate or superseded.

Keep this PR open, but it is not merge-ready: the useful bug fix is still relevant on current main, yet the branch conflicts with renamed current code, lacks real channel proof, and still has concrete message-delivery gaps.

Canonical path: Close this stale PR. The latest review rated it F, the branch still lacks merge-ready proof, and there has been no human follow-up after the durable review.

So I’m closing this here because the remaining work is already tracked in the canonical issue.

ClawSweeper PR egg

🎁 Pass real behavior proof to wake the egg and unlock a hatchable treat.

This pull request has been automatically marked as stale due to inactivity.
Please add updates or it will be closed.

ClawSweeper applied the proposed close for this PR.

• Action: closed this PR.
• Close reason: duplicate or superseded.
• Evidence: durable ClawSweeper review.