Quality Process

When to Use This Skill

Load this skill when you need to:

• Validate documentation before commits
• Run quality checks on changed files
• Audit documentation freshness
• Repair frontmatter issues
• Coordinate quality validation workflows
• Understand when and how to validate
• Invoke quality agents appropriately

Common triggers:

• Pre-commit hook reminder appears
• After creating or modifying documentation
• Before creating pull requests
• During brownfield onboarding
• When orchestrator requests validation
• User explicitly requests quality checks

Quick Reference

The quality process ensures documentation and code maintain AgentFlow standards through validation, repair, and continuous auditing. Quality is multi-dimensional:

Documentation Quality:

• Frontmatter schema compliance
• Bidirectional link verification
• Freshness checking (30-day threshold)
• Content quality assessment

Code Quality:

• TSDoc/JSDoc compliance
• Test coverage validation
• Linting and type checking
• BDD scenario alignment

Architecture Quality:

• ADR completeness and currency
• Pattern consistency
• Documentation alignment

Process Quality:

• Phase adherence
• Approval gates
• Context preservation

Rules

Critical Quality Rules

1. Never skip validation before commits unless changes are truly trivial (typos, whitespace only)
2. Always run incremental validation by default - faster and focused on recent work
3. Use full audits sparingly - before releases, weekly schedules, or on explicit request
4. Auto-repair safe issues - frontmatter fixes, stale dates, broken internal links
5. Report unsafe issues - outdated content, missing docs, architectural decisions
6. Track validation state - use .claude/validation-state.json for git-aware incremental checks
7. Respect exemptions - don't validate .claude/work/ or .gitignore files
8. Coordinate agents - docs-quality-agent for docs, code-quality-agent for code, architecture-quality-agent for ADRs

Validation Decision Rules

Run docs-quality-agent when:

• .md files changed in .claude/ or docs/
• Framework components modified (agents, skills, commands)
• Documentation content created or updated
• Pre-commit hook suggests validation

Run code-quality-agent when:

• Source code files changed
• Tests added or modified
• Dependencies updated
• Before creating pull requests

Run architecture-quality-agent when:

• Major architectural decisions made
• ADRs created or updated
• System design changed
• Patterns modified across codebase

Skip validation when:

• Pure typo fixes (no semantic changes)
• Whitespace/formatting only
• Debug log removal
• Comment updates

Workflows

1. Pre-Commit Validation Workflow

Trigger: .claude/hooks/git-commit-reminder hook blocks first commit attempt

Procedure:

1. Attempt commit
   git commit -m "message"

2. Hook blocks with reminder
   "Have you run validation?"

3. Assess changes
   git diff --name-only HEAD  # What changed?

4. Run appropriate validation
   - Docs changed? → Task tool → docs-quality-agent
   - Code changed? → npm test && npm run lint
   - Both? → Run both validations

5. Review validation output
   - Fix any errors found
   - Address warnings
   - Confirm quality standards met

6. Retry commit (second attempt allowed)
   git commit -m "message"  # Proceeds

Key points:

• First attempt = reminder (always blocks)
• Second attempt = allowed (assumes you validated)
• Don't bypass by immediately retrying without validation
• Hook is a behavioral guardrail, respect it

2. Incremental Validation Workflow

When: During development, after making changes

Procedure:

1. Invoke docs-quality-agent (incremental mode is default)

2. Agent loads validation state
   - Reads .claude/validation-state.json
   - Identifies last validated commit
   - Gets changed files since then

3. Agent validates only changed files
   - Runs validation scripts on changes
   - Checks frontmatter, links, freshness
   - Reports issues found

4. Agent performs safe auto-repairs
   - Adds missing frontmatter fields
   - Updates stale dates
   - Fixes broken internal links

5. Agent reports unsafe issues
   - Outdated content needing rewrite
   - Missing documentation
   - Broken external links

6. Agent updates validation state
   - Records current commit as validated
   - Saves timestamp
   - Tracks files validated

Benefits:

• Fast (seconds, not minutes)
• Focused on your recent work
• Catches issues immediately
• Scales with project size

3. Full Audit Workflow

When: Before releases, weekly schedule, explicit request

Procedure:

1. Invoke quality agent with full audit directive
   Task tool → docs-quality-agent with "full audit"

2. Agent validates ALL files
   - Every .md file in .claude/ and docs/
   - All code files for TSDoc
   - Complete link graph
   - All ADRs

3. Agent generates comprehensive report
   - Total files validated
   - All issues found (errors + warnings)
   - Repairs made
   - Issues requiring human review

4. Review and address findings
   - Fix critical issues first
   - Plan for warnings
   - Update stale documentation
   - Create missing docs

5. Re-run until clean
   - All validation scripts pass
   - No critical issues
   - Warnings addressed or planned

Use cases:

• Before releasing new version
• Weekly maintenance (scheduled)
• After major refactoring
• Brownfield onboarding baseline

4. Quality Agent Invocation Workflow

For documentation validation:

Task tool → docs-quality-agent

# Agent procedure:
1. Load quality-process skill (this skill)
2. Load documentation-standards skill
3. Determine scope (incremental vs full)
4. Run validation scripts
5. Perform safe repairs
6. Report results

For code validation:

Task tool → code-quality-agent

# Agent procedure:
1. Load testing-expertise skill
2. Run linting: npm run lint
3. Run type check: npm run type-check
4. Run tests: npm test
5. Check coverage: npm run test:coverage
6. Validate TSDoc compliance
7. Report results

For architecture validation:

Task tool → architecture-quality-agent

# Agent procedure:
1. Review recent changes (git log -n 10)
2. Identify architectural decisions
3. Check for corresponding ADRs
4. Validate ADR currency
5. Check pattern consistency
6. Report findings

5. Brownfield Onboarding Workflow

When: Adding AgentFlow to existing projects

Procedure:

1. Run audit command
   /docs:audit-project

2. docs-quality-agent performs brownfield audit
   - Scans for existing documentation
   - Assesses quality (None/Poor/Scattered/Good)
   - Generates inventory
   - Identifies gaps

3. Review audit report
   - Overall assessment
   - Documentation inventory
   - Recommendations

4. Follow migration plan
   - Add frontmatter to existing docs
   - Create missing README files
   - Establish bidirectional links
   - Fill documentation gaps

5. Establish baseline
   - Run full validation
   - Fix critical issues
   - Create improvement plan
   - Set up continuous validation

See: Quality Guide - Brownfield Onboarding

Decision Points

When should I run validation?

YES - Run validation:

• Before any git commit (non-trivial changes)
• After creating new documentation
• After modifying framework components
• Before creating pull requests
• Weekly scheduled audits
• After major refactoring
• During brownfield onboarding

MAYBE - Assess first:

• After code changes → Only if code has TSDoc or doc links
• After config changes → Only if functionality impacted
• After minor updates → Check if semantic changes

NO - Skip validation:

• Pure typo fixes (no semantic change)
• Whitespace/formatting only
• Debug log removal
• Timestamp updates

Which agent should I invoke?

docs-quality-agent:

• Documentation files changed (.md)
• Framework components modified
• Pre-commit hook suggests it
• User requests docs validation

code-quality-agent:

• Source code changed
• Tests added/modified
• Before creating PR
• User requests code validation

architecture-quality-agent:

• Architectural decisions made
• ADRs created/updated
• System design changed
• Pattern modifications

All agents sequentially:

• Full audit requested
• Before releases
• After major changes

Should I auto-repair or report?

Auto-repair (safe):

• Missing frontmatter fields → Add with defaults
• Stale last_checked dates → Update to current
• Malformed tags → Convert to array format
• Broken internal links → Fix if new path known
• Parent-child mismatches → Add to children array

Report for human review (unsafe):

• Outdated content → Needs rewriting
• Missing documentation → Needs creation
• Broken external links → Needs decision
• Circular relationships → Needs restructure
• Architectural drift → Needs ADR review

Common Pitfalls

1. Bypassing Pre-Commit Validation

Problem: Immediately retrying commit without running validation

Why it's bad: Hook reminder is a behavioral guardrail to prevent quality debt

Solution: Run suggested validation before second attempt

Example:

# WRONG
git commit -m "message"  # Blocked by hook
git commit -m "message"  # Immediately retry ❌

# RIGHT
git commit -m "message"  # Blocked by hook
Task tool → docs-quality-agent  # Run validation
git commit -m "message"  # Now retry ✅

2. Forgetting Documentation Freshness

Problem: Not updating last_checked even when content is current

Why it's bad: Creates false positives in stale documentation reports

Solution: Update date after review, even if no changes

Example:

# After reviewing and confirming content is current
last_checked: 2025-12-09  # Update this

3. Breaking Bidirectional Links

Problem: Adding child without updating parent, or moving files without updating references

Why it's bad: Validation fails, navigation breaks, documentation orphaned

Solution: Update both sides of relationship

Example:

# When adding new-child.md
# In new-child.md
parent: ../README.md

# ALSO update parent
# In README.md
children:
  - ./existing.md
  - ./new-child.md  # Add this

4. Using Full Audits for Everything

Problem: Running full validation when incremental would suffice

Why it's bad: Slow feedback, wastes time, reduces validation frequency

Solution: Use incremental by default, full audits sparingly

When to use each:

• Incremental: Daily development, after changes, pre-commit
• Full: Releases, weekly schedule, brownfield baseline

5. Validating Operational Context Files

Problem: Running validation on .claude/work/ files

Why it's bad: These are exempt - dynamic working files, not documentation

Solution: Skip .claude/work/ directory

Exempt locations:

• .claude/work/ - Operational context
• .gitignore files - Not tracked
• node_modules/ - Dependencies
• build/, dist/ - Generated output

Quick Reference Tables

Validation Script Quick Reference

Quality Agent Quick Reference

Validation Trigger Decision Matrix

Integration with AgentFlow Phases

Setup Phase

• Run full quality audit to establish baseline
• Validate framework documentation is current
• Ensure validation scripts operational
• Set up git hooks for continuous validation

Discovery Phase

• Validate requirement documents as created
• Check freshness of reference documentation
• Ensure Linear features properly documented
• Run incremental validation after each document

Refinement phase

• Validate BDD feature files with bdd-expertise
• Ensure mini-PRDs have proper metadata
• Verify visual specifications complete
• Run docs-quality-agent before approval

Delivery Phase

• Validate code documentation (TSDoc) during implementation
• Run tests continuously (TDD workflow)
• Check code-to-doc links
• Pre-commit validation on every commit
• Full quality check before creating PR

Comprehensive Documentation

For complete quality guidance including:

• Quality dimensions (documentation, code, architecture, process)
• Detailed validation workflows
• Quality metrics and targets
• Troubleshooting common issues
• Best practices
• Quality agent reference

See: Quality Guide (comprehensive 900+ line guide)

Related Documentation

• Documentation Standards - Metadata and linking requirements
• Documentation System Guide - Three-layer architecture and workflows
• Testing Guide - Test coverage and TDD practices
• docs-quality-agent - Documentation validation agent
• code-quality-agent - Code validation agent
• architecture-quality-agent - Architecture validation agent