mk:fix

Fixing

Unified skill for fixing issues of any complexity with structured diagnosis.

Common Workflow (Authoritative)

Bug → Mode Select → [Check Memory: standard/deep] → Scout (MANDATORY) → Diagnose
  → [quick: confirm the known cause directly | standard/deep: investigate → sequential-thinking]
  → root cause? → yes → Root-Cause Proof → Complexity → Fix ROOT CAUSE → Verify + Prevent (MANDATORY)
  → pass → Report + capture only when recurrence/salience warrants it
  → fail <3 → re-diagnose | fail 3+ → STOP

This flow is authoritative. If prose conflicts, follow the flow.

Profiles

• quick — known cause and one or two files: minimal scout, compact root-cause proof, edit, focused before/after check, and summary. No default plan, report, wiki, memory, commit prompt, or subagent fan-out.
• standard (default) — full evidence chain, root-cause proof, focused regression test, relevant suite, and review.
• deep — standard plus an approved plan when a contract or architecture changes.

HARD GATE

Do NOT propose or implement fixes before completing Steps 1-2 (Scout + Diagnose). Symptom fixes are failure. Find the cause first through structured analysis, NEVER guessing. If 3+ fix attempts fail, STOP and question the architecture — discuss with user. --quick may use the compact proof, but it never bypasses Scout, Diagnose, root-cause evidence, Verify + Prevent, or the three-failed-attempt stop.

Arguments

• --auto — Standard profile with automatic progress through diagnosis and verification. It stops at any required human gate and never self-approves.
• --review — Human-in-the-loop. Pause at each step.
• --quick — Quick profile for a known cause and ≤2 files.
• --parallel — Parallel developer agents per independent issue.
• --tdd — Force regression test BEFORE the fix (writes the .claude/session-state/tdd-mode sentinel). Without --tdd, regression tests are recommended but not gated. Useful for security-sensitive fixes where you want to prove the bug first.
• --advice — Opt-in strategic supervision for one run. At named checkpoints (GUIDE → RESCUE* → REVIEW → RECHECK*, hard cap 5) the workflow asks the athena agent to assess the situation, recommend an operational path inside the locked scope, and — after the work — review it against the evidence, with the power to return it for correction. Standard and deep profiles only. Supervision is evidence, never approval: it clears no gate, counts as no verification, and the three-failed-attempt human STOP below still fires on its own schedule. Without the flag, zero calls happen and no state is written. Contract: .claude/rules-conditional/advice-supervision-rules.md (load it only when the flag is present).