[jalay-mantys]

Problem

After changing the model inside a Claude thread (Opus 5.5 → Fable 5.2 on the same instance), pressing Stop, and then switching to another provider (Codex 6.1 Sol), the send is refused:

The previous provider is still running background agents or commands, and switching would end them. Wait for them to finish, or press Stop, then send the message again.

Nothing was running. The thread's pendingBackgroundTasks was empty, there were no subagents or background commands, and the Claude process had exited. Retrying, or pressing Stop again, gave the same refusal.

Cause

On a live model change the adapter calls set_model before it attaches the new turn. Claude Code answers with a replayed user frame:

{"type":"user","message":{"role":"user","content":"<local-command-stdout>Set model to `claude-fable-5-2`</local-command-stdout>"},"parent_tool_use_id":null,"isReplay":true}

No turn is active yet, so the frame goes to bufferWakeMessage, which treats any user frame as wake evidence. The buffered user frame makes hasPendingBackgroundWork return true. An ordinary interrupt closes the query but does not clear the wake buffer, so the provider-switch guard in Orchestrator.ts keeps refusing.

Change

bufferWakeMessage ignores replayed <local-command-stdout> user frames. They are output echoed back from a command the adapter sent itself, not background work. Turns that are active still handle these frames as before, for example /compact's "Compacted" echo.

Scope

A small bug fix with one root cause. Related: #36450 and #35240 report the same refusal. #36596 lets Stop clear background state that only the provider's probe sees; this PR removes one source of that false state, so Stop isn't needed in this case.

Verification

• Found from a real occurrence on 0.1.10-nightly.20261114.4669 (macOS desktop). The provider event log shows the Set model to replay at the model change, then query.interrupt → result: error_during_execution → query.close, with no task_started or background frames at any point. The server trace shows every later message.dispatch failing with the refusal above.
• New test in ClaudeAdapterV2.test.ts, "does not count a live model change's stdout echo as background work". It fails on main (hasPendingBackgroundWork is true) and passes with the fix.
• ClaudeAdapterV2.test.ts (425 tests), ClaudeReplayFixtures.integration.test.ts, ProviderSwitch.integration.test.ts and BackgroundWorkStop.integration.test.ts (109 tests) all pass. The compact fixtures that replay <local-command-stdout>Compacted are among them. vp run check:changed passes.
• Not checked: an end-to-end run against a real Claude CLI on a dev server.

Made with Claude Opus 5.5 in Claude Code.

[macroscopeapp]

Approvability

Verdict: Approved at 4db40b7

Macroscope's review found this PR approvable — Replayed Claude local-command output is excluded from wake buffering so changing models no longer blocks the next provider switch. The filter targets a specific echo format while preserving other wake-message handling.

^{You can add or adjust custom eligibility rules. Learn more.}

[coderabbitai]

No actionable comments were generated in the recent review.

Walkthrough

ClaudeAdapterV2 now skips replayed user frames that begin with <local-command-stdout> when buffering wake messages. A regression test checks that a replayed set_model stdout echo does not cause hasPendingBackgroundWork to return true.

Changes: Claude local-command echo filtering

Layer / File(s)	Summary
Filter replayed local-command echoes apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.ts, apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.test.ts	isClaudeLocalCommandEcho identifies replayed user frames with string content that begins with <local-command-stdout>. bufferWakeMessage skips matching frames. The regression test checks that a replayed set_model stdout echo does not report pending background work.

Estimated code review effort: 2 (Simple) | ~10 minutes

Change: Bug fix

Suggested reviewers: juliusmarminge, t3dotgg, yash-singh1

Merge Risk: Minimal · up to 4db40

Changing a Claude model and pressing Stop no longer leaves the thread looking as if it has background work, so the next provider switch should no longer be refused. The change is small and covered by a regression test. The only unchecked item is an end-to-end run against a real Claude CLI, which does not block merging.

Pre-merge checks: 4

Check name	Status	Explanation
Title check		The title names the main change: a Claude model change no longer leaves pending background work that blocks the next provider switch. The changeset adds a filter in bufferWakeMessage that ignores repl…
Description check		The description follows the template. It has Problem, Change, Scope, and Verification sections, and each contains specific content. The Problem section gives the reproduction steps and expected behavi…
Linked Issues check		Check skipped because no linked issues were found for this pull request.
Out of Scope Changes check		Check skipped because no linked issues were found for this pull request.

Full details: Title check

Explanation

The title names the main change: a Claude model change no longer leaves pending background work that blocks the next provider switch. The changeset adds a filter in bufferWakeMessage that ignores replayed <local-command-stdout> user frames, plus a regression test. The title uses the fix(server) conventional-commit form and is specific enough for a teammate scanning history.

Full details: Description check

Explanation

The description follows the template. It has Problem, Change, Scope, and Verification sections, and each contains specific content. The Problem section gives the reproduction steps and expected behavior. The Change section explains the root cause and the fix. The Scope section says this is a small bug fix and cross-references related issues. The Verification section lists focused tests, the observed results, and what was not checked. It also includes the model and harness line that the template asks for.

Finishing Touches

Generate unit tests (beta)

• Create a new PR

• Autofix · Keep fixing CodeRabbit findings and required CI, and resolving merge conflicts

Recent review info

Run configuration

• Source mode: Auto
• Repository: pingdotgg/t3code
• Branch: main
• File: .coderabbit.config.ts
• Commit: 13f89ae
• Layers: Repository TS config
• Review profile: CHILL
• Plan: Advanced
• Run ID: 77094aad-f7f5-4641-ab96-5d1d3e5da825

Commits

Reviewing files that changed from the base of the PR and between 13f89ae and 4db40b7.

Files selected for processing: 2

• apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.test.ts
• apps/server/src/orchestration-v2/Adapters/ClaudeAdapterV2.ts

Included review availability: This review used your included allowance. Your plan provides up to 10 included reviews per hour; 9 remain after this review.

[juliusmarminge]

Thanks for this. #34498 makes the same change and is the one being kept.

#34498 fixes the same replayed model-change echo being mistaken for background work. Its regression also drives a real model change through the adapter and verifies that an ordinary user frame still counts as wake evidence, which makes it the stronger version to keep.

Your report independently confirms the same cause. The broader missing Stop control and other sources of stale background state remain separate problems.

Closing as a duplicate.