Skip to content

meow:fix

Structured bug investigation and fix with auto-complexity detection, parallel exploration, and multiple workflow modes.

What This Skill Does

meow:fix is MeowKit's debugging pipeline. Instead of immediately editing code when you report a bug, it forces a structured investigation: assess complexity, choose the right workflow, find the root cause, write a regression test, then apply the minimal fix. The skill adapts its approach based on how complex the bug is — a typo gets a quick fix, while a cross-module race condition gets the full investigation treatment.

Core Capabilities

  • Auto-complexity assessment — Classifies bugs as Quick (typo, config), Standard (logic, one module), or Deep (cross-cutting, architectural)
  • Four fix modes — Autonomous (default), human-in-the-loop review, quick (for trivial issues), parallel (for multi-file problems)
  • Root cause methodology — Uses meow:investigate for systematic 5-phase debugging
  • Parallel exploration — Spawns multiple Explore subagents to verify hypotheses simultaneously
  • Regression test guarantee — Every fix includes a test that fails without the fix and passes with it
  • Scope locking — Uses meow:freeze to restrict edits to affected modules, preventing scope creep

When to Use This

Use meow:fix when...

  • You've found a bug and need it investigated properly
  • Tests are failing and you need to understand why
  • A CI/CD pipeline is broken
  • You're seeing unexpected behavior in production
  • Type errors, lint failures, or UI glitches need fixing

Don't use meow:fix when...

  • You're building a new feature → use meow:cook
  • The issue is a design/architecture concern → use architect agent

Usage

bash
# Default — autonomous mode, auto-detects complexity
/meow:fix login fails after 24 hours

# Review mode — pauses for approval at each step
/meow:fix payment processing timeout --review

# Quick mode — for trivial issues (typos, lint, config)
/meow:fix TypeScript error in auth.ts --quick

# Parallel mode — spawns agents per issue for multi-file problems
/meow:fix all failing tests in checkout module --parallel

Example Prompts

PromptComplexityMode
/meow:fix typo in README.mdQuickQuick — direct fix, no investigation
/meow:fix session token not refreshedStandardAutonomous — investigate → fix → test
/meow:fix intermittent race condition in payment queueDeepFull investigation with parallel exploration
/meow:fix CI failing on main branchStandardAutonomous — check CI logs, reproduce, fix

Fix Pipeline (7 steps)

StepWhat happensSkills used
0Mode selectionAskUserQuestion (Autonomous/HITL/Quick)
0.5Check fix memoryRead memory/lessons.md + patterns.json for known fix patterns
1Scout (mandatory)meow:scout — map files, deps, tests
2Diagnosemeow:investigate + meow:sequential-thinking
3Complexity assessmentRoute to Quick/Standard/Deep
4Fix implementationAddress ROOT CAUSE, not symptoms
5Verify + prevent (mandatory)Regression test + defense-in-depth
6Finalize + learnReport, write fix pattern to memory, docs, commit

Self-Improving Fix Loop

After every fix, Step 6 writes the fix pattern to memory/patterns.json. Next time a similar bug appears, Step 0.5 finds the pattern and fast-tracks diagnosis — turning repeated bugs into instant fixes.

Fix bug #1 → write pattern (frequency: 1)
Fix bug #2 (same class) → read pattern → skip full diagnosis → apply known fix
After 3+ fixes → pattern promoted to CLAUDE.md rule (via meow:memory)

Skill Details

Phase: 1–5
Plan-First Gate: Plans if fix affects > 2 files. Skips with --quick mode.

Gotchas

  • Fixing symptoms not root cause: Quick patch makes the test pass but underlying issue remains → Always investigate before implementing; use meow:investigate first
  • Regression in adjacent code: Fix in one module breaks an unstated dependency → Run full test suite, not just tests for the changed file
  • Test mocking hiding real failures: Mocked tests pass but real integration fails → Prefer integration tests for bug fixes; mock only external services

Released under the MIT License.