本文へ移動
cccskills
無料GitHub で公開

fix-verification

Comprehensive fix verification methodology to ensure bugs are truly fixed. Use after implementing any bug fix to verify it works and hasn't caused regressions. Keywords: verify, confirm, test, validate, check, ensure, regression, fixed

インストール方法を見る

含まれるファイル(1)

  • SKILL.md9.3 KB

SKILL.md(原文)

インストールする前に、エージェントに与えられる指示の中身を確認できます。

✅ Fix Verification Skill

Philosophy: A bug isn't fixed until it's verified. Twice.

When to Use

Use this skill when:

  • You've implemented a bug fix
  • Someone else's fix needs verification
  • Deploying a fix to production
  • Bug was high-impact and needs thorough verification
  • Previous fixes for this bug have failed

Do NOT use this skill when:

  • Just exploring code (no fix yet)
  • Bug is trivial (e.g., typo fix)
  • Running standard CI (automated verification)

Prerequisites

Before starting:

  • Fix has been implemented
  • You know the expected behavior
  • You can reproduce the original bug (or have a failing test)
  • You have access to test environment

Process

Phase 1: DIRECT VERIFICATION 🎯

Goal: Confirm the fix addresses the exact reported issue.

Steps:

  1. Recreate the Original Bug Scenario

    Exact steps that caused the bug:
    1. [step 1]
    2. [step 2]
    3. [step 3]
    
    Expected result: [what should happen]
    Previous result: [what was happening - the bug]
    
  2. Execute Test with Fix Applied

    • Follow exact same steps
    • Document actual result
    • Compare to expected
  3. Verify the FIX, Not Just Absence of Error

    ❌ "It doesn't crash anymore" (incomplete)
    ✅ "It returns the expected user object with correct fields" (complete)
    
  4. Test Multiple Times

    • Run at least 3 times
    • Note any inconsistency

Verification Checklist:

  • Original bug scenario no longer produces error
  • Expected behavior now occurs
  • Result is consistent across multiple runs
  • All variations of bug scenario work

Phase 2: EDGE CASE VERIFICATION 🔲

Goal: Ensure fix works for edge cases and variations.

Steps:

  1. Identify Edge Cases

    edge_cases:
      boundary_values:
        - Empty input
        - Maximum length input
        - Minimum valid input
        - Just over limit
        - Just under limit
      
      special_inputs:
        - Null/undefined
        - Special characters
        - Unicode/emoji
        - Numbers as strings
        - Whitespace only
      
      state_variations:
        - First time user
        - Returning user
        - Admin user
        - Concurrent users
        - High load
    
  2. Create Edge Case Matrix

    Edge CaseInputExpectedActualPass?
    Empty""Error msg
    Max length1000 charsSuccess
    Special chars"<>&"Escaped
  3. Test Each Edge Case

    • Document results
    • Note any failures

Edge Case Checklist:

  • All boundary values tested
  • Special inputs handled correctly
  • Different user states work
  • Concurrent access works (if applicable)

Phase 3: REGRESSION TESTING 🔄

Goal: Ensure fix hasn't broken anything else.

Steps:

  1. Identify Affected Areas

    affected_areas:
      directly_affected:
        - The fixed function/component
        - Its callers
        - Its dependencies
      
      indirectly_affected:
        - Related features
        - Shared utilities used
        - Configuration changes
    
  2. Run Targeted Tests

    # Run tests for affected module
    npm test -- --grep "auth"
    
    # Run integration tests
    npm run test:integration
    
  3. Run Full Test Suite

    # Ensure no regressions anywhere
    npm test
    
  4. Manual Smoke Test

    • Test main user flows
    • Pay attention to anything that feels different

Regression Checklist:

  • All unit tests pass
  • All integration tests pass
  • Smoke test passes
  • No new warnings in logs
  • Performance not degraded

Phase 4: NEGATIVE TESTING 🚫

Goal: Verify error handling and failure modes.

Steps:

  1. Test Invalid Inputs

    • What happens with bad data?
    • Are errors handled gracefully?
    • Are error messages helpful?
  2. Test Failure Scenarios

    failure_scenarios:
      - Database connection lost
      - API timeout
      - Invalid credentials
      - Missing permissions
      - Disk full
    
  3. Test Recovery

    • Does system recover after failure?
    • Is data consistent after recovery?

Negative Testing Checklist:

  • Invalid inputs produce clear errors
  • System fails gracefully
  • Recovery works correctly
  • No security information leaked in errors

Phase 5: ENVIRONMENT VERIFICATION 🌍

Goal: Ensure fix works across all environments.

Steps:

  1. Test Across Environments

    EnvironmentTestedResultNotes
    Local dev
    CI/CD
    Staging
    Production
  2. Test Across Configurations

    • Different browsers (if applicable)
    • Different OS (if applicable)
    • Different database sizes
    • Different load levels
  3. Test Across User Types

    • Regular users
    • Admin users
    • New users
    • Users with existing data

Environment Checklist:

  • Works in development
  • Works in staging
  • Works in production (or production-like)
  • Works across configurations

Phase 6: DOCUMENTATION & SIGN-OFF 📝

Goal: Document verification and get sign-off.

Steps:

  1. Complete Verification Report

    ## Fix Verification Report
    
    **Bug ID:** [ID]
    **Fix Description:** [brief description]
    **Verified By:** [name]
    **Date:** [date]
    
    ### Direct Verification
    - [x] Original bug no longer occurs
    - [x] Expected behavior confirmed
    
    ### Edge Cases
    - [x] [edge case 1] - PASS
    - [x] [edge case 2] - PASS
    
    ### Regression Testing
    - [x] Unit tests: 100% pass
    - [x] Integration tests: 100% pass
    - [x] Smoke test: PASS
    
    ### Environment Testing
    - [x] Local: PASS
    - [x] Staging: PASS
    - [x] Production: [pending/pass]
    
    ### Sign-off
    - [ ] Developer
    - [ ] QA (if applicable)
    - [ ] Stakeholder (for critical bugs)
    
  2. Update Bug Tracking

    • Change status to "Verified Fixed"
    • Add verification notes
    • Link to any new tests added
  3. Add/Update Tests

    • Ensure test exists for this bug
    • Test should fail without fix, pass with fix

Quick Verification Checklist

For simpler bugs, use this shortened checklist:

## Quick Verification

- [ ] Original bug scenario fixed
- [ ] Edge cases work
- [ ] All tests pass
- [ ] No new warnings/errors
- [ ] Reviewed by another person (if critical)

Verification Decision Tree

Fix Implemented
      │
      ▼
┌─────────────────┐
│ Does original   │
│ scenario work?  │
└────────┬────────┘
    YES  │  NO → Back to debugging
         ▼
┌─────────────────┐
│ Edge cases      │
│ work?           │
└────────┬────────┘
    YES  │  NO → Expand fix
         ▼
┌─────────────────┐
│ All tests       │
│ pass?           │
└────────┬────────┘
    YES  │  NO → Fix regressions
         ▼
┌─────────────────┐
│ Works in all    │
│ environments?   │
└────────┬────────┘
    YES  │  NO → Environment-specific fix
         ▼
   VERIFIED ✅

Guidelines

DO ✅

  • Verify the fix, not just absence of error
  • Test edge cases thoroughly
  • Document your verification
  • Run the full test suite
  • Have someone else verify critical fixes

DON'T ❌

  • Assume fix works because code looks right
  • Skip edge case testing
  • Forget to check for regressions
  • Test only in development environment
  • Skip documentation

Common Pitfalls

"Works on My Machine"

Problem: Fix works locally but fails elsewhere.

Prevention:

  • Always test in staging
  • Use same data/config as production
  • Test with production-like load

"Fixed the Symptom, Not the Bug"

Problem: Error is gone but behavior is still wrong.

Prevention:

  • Verify POSITIVE behavior, not just absence of error
  • Compare to expected behavior documentation
  • Test the complete user flow

"Created a New Bug"

Problem: Fix introduced a regression.

Prevention:

  • Always run full test suite
  • Manual smoke test of related features
  • Review changes with fresh eyes

Success Criteria

Before marking bug as fixed:

  • Original issue verified fixed
  • All edge cases pass
  • No regression in tests
  • Works in all environments
  • Verification documented
  • Test added to prevent recurrence

Related Skills

  • skills/kilo-kit/debugging/systematic/ - For finding the bug
  • skills/kilo-kit/debugging/root-cause/ - For understanding why it happened
  • skills/kilo-kit/quality/testing/ - For writing better tests

Fix Verification Skill v1.0.0 — Verified twice, deployed once

レビュー

まだレビューはありません。使ってみた感想をお寄せください。

同じリポジトリのスキル

概要と使いどころ

2d-games

無料

2D game development principles. Sprites, tilemaps, physics, camera.

日本語の概要は準備中です。原文の説明を表示しています。

VoDaiLocz/kilo-kit-mcp272026年9月13日 更新

3d-games

無料

3D game development principles. Rendering, shaders, physics, cameras.

日本語の概要は準備中です。原文の説明を表示しています。

VoDaiLocz/kilo-kit-mcp272026年9月13日 更新

aesthetic

無料

Create aesthetically beautiful interfaces following proven design principles. Use when building UI/UX, analyzing designs from inspiration sites, generating design images with ai-multimodal, implementing visual hierarchy and color theory, adding micro-interactions, or creating design documentation. Includes workflows for capturing and analyzing inspiration screenshots with chrome-devtools and ai-multimodal, iterative design image generation until aesthetic standards are met, and comprehensive design system guidance covering BEAUTIFUL (aesthetic principles), RIGHT (functionality/accessibility), SATISFYING (micro-interactions), and PEAK (storytelling) stages. Integrates with chrome-devtools, ai-multimodal, media-processing, ui-styling, and web-frameworks skills.

日本語の概要は準備中です。原文の説明を表示しています。

VoDaiLocz/kilo-kit-mcp272026年9月13日 更新

Use when implementing or managing persistent, hierarchical memory systems for AI agents. Covers cross-session state, fact supersession, and self-managed memory tools to enable long-term recall and adaptive agent behavior.

日本語の概要は準備中です。原文の説明を表示しています。

VoDaiLocz/kilo-kit-mcp272026年9月13日 更新

Use when monitoring, tracing, or debugging agentic workflows in production. Keywords: observability, tracing, OpenTelemetry, Langfuse, latency, token cost, loop detection, telemetry.

日本語の概要は準備中です。原文の説明を表示しています。

VoDaiLocz/kilo-kit-mcp272026年9月13日 更新

Use when building self-correcting retrieval systems for AI agents. Keywords: RAG, retrieval, Corrective RAG, Self-RAG, query decomposition, reranking, hallucination, grounding.

日本語の概要は準備中です。原文の説明を表示しています。

VoDaiLocz/kilo-kit-mcp272026年9月13日 更新

VoDaiLocz のスキルをすべて見る

このスキルの問題を報告する