fabrik-review-comment

Read the current stage output from .fabrik-context/stage-Review.md, interpret each comment as a fix, dismiss, defer, or clarify decision, apply minimal targeted changes, then commit and push. Write results to stdout so the engine posts the formatted comment.

Use the File and Line headers to navigate directly to the location, read the diff hunk first, and group comments sharing the same thread ID. For more context, run gh api /repos/{owner}/{repo}/pulls/{pr_number}/comments.

No. Posting directly bypasses the engine's comment formatting, creates duplicates, and triggers a self-review loop. Write output to stdout only; resolving review threads via the GraphQL resolveReviewThread mutation is the permitted exception.

Claude Code's background bash detaches processes via setsid, so they survive the stage and become orphaned processes holding ports. Prefer one-shot builds, bracket servers with trap-based teardown in a single command, or bound them with a timeout.

Follow the No-Op Contract: change nothing and state that no actionable findings exist. Inventing a speculative fix can trigger fresh bot reviews, bypass deduplication, and consume extra review cycles.

Only after a code change addressing the finding, or after stating an explicit risk rationale covering reachability, trigger conditions, and existing mitigations. Citing spec conformance or prior review provenance alone does not justify resolution.