Repository navigation
Tell an auto author to relay a failed review's error, not guess at it - #25
Merged
Merged
Conversation
When a managed review failed because of a reviewer override, an author blamed the user's Codex config and promised a retry would work; both were wrong. The instructions and the author's system prompt now say: show the error as printed, don't guess at a cause it doesn't name, and when it names --reviewer-model or --reviewer-effort, ask the user to restart with another value instead of retrying. The instructions also no longer say the partner always runs in another terminal. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012Y3kg5hjT2zSZM9qUEEinM
…e reminder Every failed round reminds the author that retries keep the reviewer settings, so naming an override doesn't mean it caused the failure. Only reviewer output saying the model or effort is invalid or unsupported calls for a restart; a timeout still retries. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012Y3kg5hjT2zSZM9qUEEinM
An xhigh review found the "invalid or unsupported" condition too narrow (a missing model is often "not found" or "does not exist"), timeouts not always worth retrying, the two copies of the rule already out of step, and no cap on retries, all because the error told the author to stop whenever an override was set and the prompts told it to ignore that. The error now only states facts: what pair wait keeps, and that changing it takes pair stop and a new pair start. The author retries once; if the same failure comes back, it stops and asks the user, naming any reviewer setting the error says a retry keeps. The rule lives in one constant used by the author's system prompt, and a test keeps prompts/pair.md saying the same. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012Y3kg5hjT2zSZM9qUEEinM
The author's send already adds that the request remains on disk, so the failure no longer says Request saved too, and 'changing that' fits one override or two. Co-Authored-By: Claude Opus 5.5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_012Y3kg5hjT2zSZM9qUEEinM
Merged
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
This morning a managed review failed because of
--reviewer-model GPT-6.1-Sol. The auto author blamed the user's Codex config, and said apair waitretry would work once that was fixed. Both were wrong. Since 0.7.0 the error itself names the override and says a retry keeps it. This makes the author's instructions match.prompts/pair.md, Auto mode: if a managed review fails, show the user the error as printed, including the quoted reviewer output, and don't guess at a cause it doesn't name. If it names--reviewer-modelor--reviewer-effort, tell the user topair stopand start again with another value;pair start --resumekeeps the conversation. Otherwisepair waitretries.src/auto.tsgets the same guidance in one sentence.This changes the instructions in every worktree. They refresh on the next
sendorwait, as with any release. No test pins this wording, and 111 tests pass.🤖 Generated with Claude Code
https://claude.ai/code/session_012Y3kg5hjT2zSZM9qUEEinM