fix

Arguments (optional):

• First positional (integer 0–100) → threshold (default 60). The Phase 8 fix gate: confirmed_mechanical/partial/regression findings with score_phase4 >= threshold are eligible. /adamsreview:fix 80 excludes moderate-strength findings from the run.
• --granular-commits → one commit per surviving fix group. Default is one combined commit for all survivors.

Read fragments/_prelude-shared.md before proceeding — it lists rules that apply to every phase below (sub-agent return handling, helper-script error-as-prompt).

Execution overview

This command orchestrates Phases 7, 7.5, 8, and 9 in order. Each phase is defined in a fragment under fragments/NN-<name>.md. At each phase boundary below, read the named fragment with the Read tool and execute the instructions inside before proceeding to the next phase.

Phase 7.5 (auto-recommendation preflight) lives at the end of fragments/08-fix-loader.md. When that fragment populates Phase 5.5's auto_fix_hint on findings, Phase 7.5 surfaces them in a batch-confirm UI before Phase 8 dispatch — apply-all / review-per- finding / skip / cancel. On the apply-all and review paths, the user choices are committed via human_confirmation so Phase 8's gate picks them up via the existing bypass (no Phase 8 wiring change needed). On skip, Phase 8 runs against originally-eligible findings only. On cancel, :fix aborts and any stash from 7.5's clean-tree gate is restored.

Before you start, build a TaskList that mirrors the phases below (one task per phase, plus one for argument parsing). Mark each in_progress when you start it and completed when you finish.

attempted is the transient recovery anchor. Between Phase 8 completing and Phase 9e writing fix_attempts, findings sit in current_state=attempted. If the run is interrupted there — or if Phase 9.pre detects a touched-file overlap and the reviewer chooses abort (default) or inspect — the next /adamsreview:fix invocation's Phase 7 step 4 hard abort catches the leftover attempted state and gives the user a deterministic recovery prompt. Never clean up attempted state silently; that's the user's call. (Reconcile does NOT leave findings at attempted — it commits or reverts just like a non-reconciled run.)

In Phase 9e, state transitions + fix_attempts append + schema validate

• render all run BEFORE any git push or artifact-publish.sh call. Push or publish failure never leaves the artifact out of sync with git.

Phase 9b reverts any fix group whose findings Phase 9a classified as regression, before Phase 9c commits. Surviving groups' files stage explicitly by name (never git add -A); regression-group files are restored with git checkout -- (for modified files) or rm -f (for created files).

Push failure, publish failure, stash-pop conflict — each gets logged to trace.md with a specific tag; the terminal block proceeds; the FIRST failure is surfaced to the user at the end. The artifact's state is already persisted so the user can re-run cleanly.

If a phase genuinely cannot run, mark the task completed with a one-line trace.md note and move on. Phase 9e's terminal cleanup runs unconditionally — every run finishes with artifact state consistent with what actually happened on disk.