fix — Fix-Review Loop

fix — Fix-Review Loop

Runtime syntax: invoke the plugin copy through the active runtime's owner-qualified skill selector; use a bare selector only when intentionally selecting a user-tier copy (a separately installed personal copy; Genie no longer seeds this tier). Cross-skill prose below uses bare names as portable semantic routes; the orchestrator resolves the selector for the active runtime.

Resolve FIX-FIRST gaps from review: dispatch a fix subagent, re-review, repeat up to 2 loops, then diagnose and route any unresolved failure.

When to Use

• review returned a FIX-FIRST verdict with CRITICAL or HIGH gaps
• Orchestrator hands off unresolved gaps after execution review

Flow

1. Parse and diagnose gaps: severity, files, failing checks from the FIX-FIRST verdict. If the evidence is overdesigned-plan, stop and return to brainstorm/wish; removing optional machinery is a plan correction, not a code-fix attempt.
2. Dispatch fixer: native delegation surface → fix subagent, briefed with the gap list, the original wish criteria, and any trace diagnosis.
3. Re-review: native delegation surface → a separate reviewer subagent (never the fixer) running review on the same pipeline.
4. Evaluate verdict:

5. Route the diagnosis: report the remaining gaps with exact files, failing checks, cause class, and corrective route; the group's task stays in_progress.

Dispatch

Fix and re-review are separate native-dispatch dispatches — never combined in one subagent, and the re-reviewer is never the fixer. Subagents notify on completion — no polling. Follow-ups to a running fixer go through native follow-up messaging.

The fixer's brief must carry: the severity-tagged gaps (file:line), the original wish acceptance criteria, the validation command(s) to re-run, and stop conditions — fix only the listed gaps; report blocked rather than expand scope.

Escalation Diagnosis

Use this policy before any model or effort change; keep this contract identical in fix, review, and work.