source-command-fix

Use this skill when the user asks to run the migrated source command fix.

Command Template

/fix — Structured Bug-Fix Pipeline

Usage

/fix [--auto | --quick | --review | --parallel] <bug description>

Auto-detect mode if no flag given based on description:

• Review (default) — pause for approval at each step
• Auto (--auto) — auto-approve fixes if confidence ≥ 9.5 with 0 critical
• Quick (--quick) — fast cycle for trivial issues (lint, type errors, build errors); skip review + docs
• Parallel (--parallel) — scout 3 areas simultaneously for complex multi-file bugs

---

Step 0 — Scope Detect

# Scope (Step 0):
#   Description: {what the user said}
#   Quick? → {yes/no — reason}
#   Mode: {Review | Auto | Quick | Parallel}

If --quick or description is clearly a build/compiler/lint error: skip scout, go straight to Step 2 with the error message as input.

For performance-related bugs:

• prefer root-cause fixes at query shape/data flow level before infra-level tuning,
• preserve API contract compatibility by default,
• and return explicit verification evidence (tests plus at least one measurable signal) before marking fixed.

---

Step 1 — Scout

Spawn the scout agent with the bug description:

• Greps for error patterns in logs and stack traces
• Reads affected source files and maps dependencies
• Checks recent git changes for related commits

// spawning scout agent
//
// Evidence:
//   Error pattern: NullReferenceException at auth.ts:45
//   Affected files: auth.ts, session.ts
//   Recent change: commit a3f2b1 modified auth.ts (2h ago)

---

Step 2 — Diagnose

Spawn the debugger agent with the scout evidence report:

• Forms 2–3 hypotheses from the evidence
• Confirms or rejects each against the codebase
• Applies the minimal fix at the confirmed root cause location
• Returns debug report with root cause + fix applied

// spawning debugger agent
//
// Hypothesis A: null check missing in auth.ts:45 → CONFIRMED ✓
// Hypothesis B: race condition in session init   → REJECTED ✗
//
// Root cause: missing null guard on req.user before .validate()
// Fix applied: auth.ts:45 — added null guard before validate()
// Severity: HIGH | Scope: 1 file

---

Step 3 — Review (code-reviewer)

Spawn code-reviewer (skip in --quick):

• Correctness — does the fix address the root cause?
• Security — no new vulnerabilities introduced?
• Regressions — does anything else break?
• Code quality — follows project standards?

// spawning code-reviewer agent
//
// Correctness: ✓ Root cause addressed
// Security:    ✓ No new vulnerabilities
// Regressions: ✓ No side effects
// Score: 9.8/10 — APPROVED

Auto mode: auto-approve if score ≥ 9.5, 0 critical. Up to 3 fix/re-review cycles, then escalate. Review mode: always pause for approval with full findings. Quick mode: skipped entirely.

---

Step 4 — Finalize (MANDATORY)

Always required — fix is incomplete without git-manager:

project-manager (skip in --quick) — syncs plan progress if bug was tracked
docs-manager (skip in --quick) — updates docs if the fix changes a public contract
git-manager (always) — conventional commit + asks to push

// MANDATORY finalize:
// project-manager → task marked resolved (if tracked)
// docs-manager    → no doc changes needed
// git-manager     → fix(auth): add null guard on req.user before validate
//                → Push to remote? [y/N]

---

Agents

Agent	Step	Modes
scout	1 — evidence gathering	All (skip if --quick + obvious error)
debugger	2 — root cause + apply fix	All
code-reviewer	3 — quality check	Skip in --quick
project-manager	4 — sync plan/tasks	Skip in --quick
docs-manager	4 — update docs	Skip in --quick
git-manager	4 — commit + push	Always (mandatory)

---

Integration

• /plan → /cook → /fix — fix regressions found after cooking
• /code-review — standalone review without the full fix pipeline