Imported from majiayu000/claude-skill-registry-data (data/apply-qa-fixes/SKILL.md). Install upstream with npx skills add majiayu000/claude-skill-registry-data --skill apply-qa-fixes. Copyright stays with the author.

Apply QA Fixes

Systematically consume QA quality gate outputs and apply prioritized fixes for issues identified during quality review.

Purpose

After Quinn's quality review creates a quality gate with CONCERNS or FAIL status, this skill enables James to systematically consume the findings and apply fixes in a deterministic, prioritized manner.

Core Principles:

• Deterministic prioritization - Always same priority order (high severity → NFR failures → coverage gaps)
• Minimal changes - Fix exactly what's in the gate, don't over-engineer
• Validation loop - Always run tests after fixes, iterate until clean
• Observability - Track all fixes applied, tests added, coverage improvement

Integration:

• Input: Quality gate YAML + assessment reports from Quinn
• Processing: Parse findings → Build fix plan → Apply fixes → Validate
• Output: Fixed code + updated task file + telemetry

When to Use This Skill

This skill should be used when:

• Quality gate status is CONCERNS or FAIL
• Quinn has identified specific issues requiring fixes
• Task is in Review status with completed implementation

This skill should NOT be used when:

• Quality gate is PASS (no fixes needed)
• Quality gate is WAIVED (intentionally accepted)
• Task implementation is incomplete
• Issues require architectural changes (escalate to user)

Prerequisites

• Task file exists at .claude/tasks/{task-id}.md
• Quality gate exists at .claude/quality/gates/{task-id}-gate-{date}.yaml
• Assessment reports exist in .claude/quality/assessments/
• bmad-commands skill available
• Tests can be executed

Workflow

Step 0: Load Configuration and Quality Gate

Purpose: Load project configuration, locate task file, find latest quality gate, verify readiness.

Actions:

1. Load Configuration:

   python .claude/skills/bmad-commands/scripts/read_file.py \
     --path .claude/config.yaml \
     --output json

   Parse configuration to extract:

   • Quality directory paths
   • Test framework
   • Coverage thresholds

2. Load Task File:

   python .claude/skills/bmad-commands/scripts/read_file.py \
     --path .claude/tasks/{task_id}.md \
     --output json

   Verify task is ready for fixes:

   • Status is "Review" or "InProgress"
   • Implementation Record section exists
   • Quality Review section exists (from Quinn)

3. Locate Latest Quality Gate:

   Search .claude/quality/gates/ for latest gate file:

   • Pattern: {task-id}-gate-{date}.yaml
   • Select most recent by date
   • Verify gate status is CONCERNS or FAIL

4. Load Quality Gate:

   python .claude/skills/bmad-commands/scripts/read_file.py \
     --path .claude/quality/gates/{gate_file} \
     --output json

   Parse gate YAML structure:

   • decision - Gate status (PASS/CONCERNS/FAIL/WAIVED)
   • overall_score - Quality score (0-100)
   • dimensions - Quality dimension scores
   • top_issues - High/medium/low severity issues
   • nfr_validation - NFR assessment results
   • trace - Traceability gaps
   • test_design - Test coverage gaps
   • action_items - Recommended actions

Halt if:

• Configuration missing
• Task file not found
• Quality gate not found
• Gate status is PASS or WAIVED (no fixes needed)

See: references/templates.md for gate schema and configuration format

Step 1: Parse QA Findings

Purpose: Extract all issues from quality gate and assessment reports.

Actions:

1. Parse Top Issues from Gate:

   From gate.yaml:

   top_issues:
     - id: "SEC-001"
       category: "security"
       severity: "high"
       description: "Missing input validation on user endpoint"
       location: "src/api/user.ts:45"
       recommendation: "Add Joi validation schema"

   Categorize by severity:

   • High severity - Security vulnerabilities, data loss risks, critical bugs
   • Medium severity - Performance issues, reliability concerns, code quality
   • Low severity - Minor improvements, style issues, warnings

2. Parse NFR Failures:

   From gate.nfr_validation:

   nfr_validation:
     security:
       status: "FAIL"
       score: 45
       concerns:
         - "Missing SQL parameterization (src/db/queries.ts:23)"
         - "No rate limiting on API endpoints"

   Extract:

   • FAIL statuses (must fix)
   • CONCERNS (should fix)
   • Specific locations and recommendations

3. Parse Test Coverage Gaps:

   From gate.test_design.coverage_gaps:

   coverage_gaps:
     - ac_id: "AC2"
       description: "User authentication flow"
       missing_tests:
         - "Should reject invalid credentials"
         - "Should handle expired tokens"

4. Parse Traceability Gaps:

   From gate.trace.ac_gaps:

   ac_gaps:
     - ac_id: "AC4"
       description: "Secure password storage"
       coverage: "partial"
       missing: ["bcrypt hashing not implemented"]

Outputs:

• high_severity_issues[] - List of high severity issues with locations
• medium_severity_issues[] - List of medium severity issues
• low_severity_issues[] - List of low severity issues
• nfr_failures[] - NFR FAIL statuses with fixes needed
• nfr_concerns[] - NFR CONCERNS with recommended fixes
• coverage_gaps[] - Test scenarios to add
• trace_gaps[] - Uncovered acceptance criteria

See: references/templates.md for full gate schema and parsing examples

Step 2: Build Fix Plan (Priority Order)

Purpose: Create deterministic, prioritized fix plan ensuring critical issues addressed first.

Priority Rules:

1. Priority 1: High Severity Issues (security, data loss, critical bugs)
2. Priority 2: NFR Failures (status = FAIL)
3. Priority 3: Test Coverage Gaps (P0 scenarios from test-design)
4. Priority 4: NFR Concerns (status = CONCERNS)
5. Priority 5: Traceability Gaps (uncovered ACs)
6. Priority 6: Medium Severity Issues
7. Priority 7: Low Severity Issues

Within Same Priority:

• Security > Performance > Reliability > Maintainability
• Alphabetical by issue ID

Actions:

1. Build Priority List:

   fix_plan = [
     // Priority 1: High severity
     { priority: 1, issue_id: "SEC-001", type: "security", ... },
     { priority: 1, issue_id: "SEC-002", type: "security", ... },
   
     // Priority 2: NFR failures
     { priority: 2, issue_id: "NFR-SEC-001", type: "nfr_fail", ... },
   
     // Priority 3: Coverage gaps
     { priority: 3, issue_id: "TEST-001", type: "coverage_gap", ac_id: "AC2", ... },
   
     // ... continue for all priorities
   ]

2. Calculate Impact:

   For each fix, estimate:

   • Files to modify
   • Diff lines expected
   • Tests to add
   • Total effort

3. Check Guardrails:

   Verify fix plan doesn't exceed James's guardrails:

   • Max 10 files per fix session
   • Max 800 diff lines
   • If exceeded, split into multiple fix sessions or escalate

4. Present Fix Plan to User:

   === Fix Plan ===
   
   Total Issues: 15
   
   Priority 1 (High Severity): 3 issues
   - SEC-001: Missing input validation (src/api/user.ts)
   - SEC-002: SQL injection risk (src/db/queries.ts)
   - PERF-001: N+1 query issue (src/services/data.ts)
   
   Priority 2 (NFR Failures): 2 issues
   - NFR-SEC-001: No rate limiting
   - NFR-REL-001: Missing error logging
   
   Priority 3 (Coverage Gaps): 5 tests to add
   - AC2: User authentication flow (2 tests)
   - AC4: Secure password storage (3 tests)
   
   Estimated Impact:
   - Files to modify: 8
   - Tests to add: 5
   - Expected diff: ~600 lines
   
   Proceed with fixes? (y/n)
   

Outputs:

• fix_plan[] - Ordered list of fixes to apply
• estimated_impact{} - Files, lines, tests
• guardrail_check{} - Pass/fail status

See: references/priority-rules.md for detailed priority logic and references/examples.md for fix plan examples

Step 3: Apply Fixes

Purpose: Apply fixes from plan using bmad-commands primitives and Claude Code tools.

Actions:

For each fix in fix_plan (in priority order):

1. Load Affected Files:

   Use Claude Code Read tool to load files needing modification:

   Read: src/api/user.ts
   Read: tests/api/user.test.ts
   

2. Apply Code Fix:

   Use Claude Code Edit tool to apply fix:

   Edit: src/api/user.ts
   Old: const user = await db.query(`SELECT * FROM users WHERE id = ${userId}`)
   New: const user = await db.query('SELECT * FROM users WHERE id = ?', [userId])
   

   Principles:

   • Minimal changes (fix exactly what's specified)
   • Follow project patterns (maintain consistency)
   • Add comments where needed (explain security fixes)
   • No over-engineering (don't refactor unrelated code)

3. Add Missing Tests:

   Use Claude Code Write tool to add test files:

   Write: tests/api/user-auth.test.ts
   Content: [Test scenarios from coverage_gaps]
   

   Use Edit tool to add tests to existing files:

   Edit: tests/api/user.test.ts
   Add tests for:
   - Should reject invalid credentials
   - Should handle expired tokens
   

4. Track Progress:

   Update todo list as fixes applied:

   ✅ Fix 1/15: Added input validation (SEC-001)
   ✅ Fix 2/15: Added SQL parameterization (SEC-002)
   ⏳ Fix 3/15: Adding rate limiting middleware...
   

Error Handling:

If fix fails:

• Log error with details
• Skip to next fix
• Report all failures at end
• Allow user to review and retry

Outputs:

• fixes_applied[] - List of successfully applied fixes
• fixes_failed[] - List of failed fixes with errors
• files_modified[] - All files changed
• tests_added - Count of new tests

See: references/examples.md for fix application examples

Step 4: Validate

Purpose: Run tests and lint to verify fixes don't introduce regressions.

Actions:

1. Run Lint:

   # Lint check (exact command depends on project)
   npm run lint
   # or
   eslint src/

   Capture output:

   • Errors count
   • Warnings count
   • Files with issues

2. Run Tests:

   python .claude/skills/bmad-commands/scripts/run_tests.py \
     --path . \
     --framework auto \
     --output json

   Parse results:

   {
     "success": true,
     "outputs": {
       "passed": true,
       "total_tests": 45,
       "tests_passed": 45,
       "tests_failed": 0,
       "coverage_percent": 89
     }
   }

3. Check Coverage Improvement:

   Compare before/after coverage:

   • Before: 82% (from quality gate)
   • After: 89%
   • Improvement: +7%

4. Iterate if Needed:

   If tests fail:

   • Identify failing tests
   • Debug issues
   • Apply additional fixes
   • Re-run tests
   • Maximum 3 iterations before escalating

Halt if:

• Tests still failing after 3 iterations
• Lint errors can't be resolved
• Coverage decreased (regression)

Outputs:

• lint_passed - Boolean
• tests_passed - Boolean
• coverage_before - Number
• coverage_after - Number
• coverage_improvement - Number
• validation_clean - Boolean

See: references/templates.md for validation report format

Step 5: Update Task File

Purpose: Update task Implementation Record with fix details.

STRICT PERMISSIONS:

AUTHORIZED to update:

• Implementation Record section (what fixed, why, how)
• File List (add/modified/deleted files)
• Status (only if re-review needed: InProgress → Review)

NOT AUTHORIZED to update:

• Objective
• Acceptance Criteria
• Context
• Tasks
• Quality Review section (that's Quinn's territory)

Actions:

1. Load Current Task File:

   python .claude/skills/bmad-commands/scripts/read_file.py \
     --path .claude/tasks/{task_id}.md \
     --output json

2. Generate Fix Summary:

   ### QA Fixes Applied (2025-01-15)
   
   **Quality Gate:** .claude/quality/gates/{task-id}-gate-{date}.yaml
   **Status Before:** CONCERNS (72/100)
   **Status After:** Ready for re-review
   
   **Fixes Applied (15 total):**
   
   **Priority 1 (High Severity):**
   1. SEC-001: Added input validation (src/api/user.ts:45)
   2. SEC-002: Added SQL parameterization (src/db/queries.ts:23)
   3. PERF-001: Fixed N+1 query (src/services/data.ts:67)
   
   **Priority 2 (NFR Failures):**
   4. NFR-SEC-001: Added rate limiting middleware
   5. NFR-REL-001: Added error logging and monitoring
   
   **Priority 3 (Coverage Gaps):**
   6-10. Added 5 missing tests for AC2, AC4
   
   **Priority 4-7:**
   11-15. Addressed medium/low severity issues
   
   **Validation Results:**
   - ✅ Lint: 0 problems
   - ✅ Tests: 45/45 passed
   - ✅ Coverage: 82% → 89% (+7%)
   
   **Files Modified:**
   - src/api/user.ts
   - src/db/queries.ts
   - src/services/data.ts
   - src/middleware/rate-limit.ts (new)
   - tests/api/user.test.ts
   - tests/api/user-auth.test.ts (new)
   - tests/db/queries.test.ts
   
   **Next Steps:**
   - Ready for Quinn re-review: @quinn *review {task-id}

3. Update Implementation Record:

   Use Edit tool to append fix summary to Implementation Record section:

   Edit: .claude/tasks/{task-id}.md
   Section: Implementation Record
   Action: Append QA fix summary
   

4. Update File List:

   Add newly created/modified files to task file list.

5. Update Status (if needed):

   If fixes require re-review:

   • Status: Review (keep in review)
   • Or Status: InProgress → Review (if was moved out of review)

Outputs:

• task_updated - Boolean
• fix_summary - Markdown text

See: references/templates.md for task update template

Step 6: Emit Telemetry

Purpose: Track metrics for observability and learning.

Actions:

Emit structured telemetry:

{
  "event": "skill.apply-qa-fixes.completed",
  "timestamp": "2025-01-15T10:30:00Z",
  "data": {
    "task_id": "task-001",
    "quality_gate": ".claude/quality/gates/task-001-gate-2025-01-15.yaml",
    "fixes_count": 15,
    "fixes_applied": {
      "high_severity": 3,
      "nfr_failures": 2,
      "coverage_gaps": 5,
      "nfr_concerns": 2,
      "medium_severity": 2,
      "low_severity": 1
    },
    "tests_added": 5,
    "coverage_before": 82,
    "coverage_after": 89,
    "coverage_improvement": 7,
    "files_modified": 7,
    "validation_passed": true,
    "duration_ms": 180000,
    "iterations": 1,
    "fixes_failed": 0
  }
}

Metrics to Track:

• Fixes count by priority
• Tests added
• Coverage improvement
• Files modified
• Validation success
• Duration
• Iterations needed
• Failed fixes

Use Cases:

• Track most common QA issues
• Identify patterns in fixes
• Measure fix effectiveness
• Monitor coverage improvement trends
• Optimize fix prioritization

Execution Complete

Skill complete when:

• ✅ Quality gate loaded and parsed
• ✅ Fix plan built with deterministic prioritization
• ✅ All prioritized fixes applied (or documented failures)
• ✅ Tests added for coverage gaps
• ✅ Validation clean (lint + tests pass)
• ✅ Task file updated with fix summary
• ✅ Telemetry emitted

Integration with James

James Routing:

# In james-developer-v2.md

command: "*apply-qa-fixes"

routing_logic:
  - condition: "high_severity_count > 5 OR nfr_failures > 2"
    guardrails:
      max_files: 10
      max_diff_lines: 800
      require_user_confirmation: true

  - condition: "high_severity_count > 0 OR nfr_failures > 0"
    guardrails:
      max_files: 7
      max_diff_lines: 600

  - condition: "default"
    guardrails:
      max_files: 5
      max_diff_lines: 400

Usage:

# After Quinn review creates gate with CONCERNS/FAIL
@james *apply-qa-fixes task-001

# James:
# ✅ Quality gate loaded: CONCERNS (3 high, 2 medium issues)
# ✅ Fix plan created (5 fixes prioritized)
# ⏳ Applying fixes...
# ✅ All fixes applied
# ✅ Validation passed
# ✅ Task updated
#
# Ready for re-review: @quinn *review task-001

Best Practices

1. Always validate after fixes - Don't skip test runs
2. Follow priority rules strictly - Ensures consistency
3. Minimal changes only - Fix what's specified, no more
4. Document all fixes - Update Implementation Record
5. Track failures - Log any fixes that can't be applied
6. Re-review after fixes - Always have Quinn re-assess

When to Escalate

Escalate to user when:

• Fixes exceed guardrails (>10 files, >800 lines)
• Fixes require architectural changes
• Tests fail after 3 iterations
• Security concerns arise during fixes
• Breaking API changes required
• Conflicting fixes (can't satisfy all issues)

References

• references/templates.md - Gate schema, task update template, output formats
• references/priority-rules.md - Detailed priority logic and decision rules
• references/examples.md - Fix application examples and usage scenarios

Part of BMAD Enhanced Development Suite - Integrates with Quinn (Quality) for complete feedback loop