Claude Code Quality Validation Checklists

Created: 2025-10-21 Source: PR #164 claude-code-expert (your-org/da-agent-hub) Purpose: Systematic quality validation for Claude Code configuration artifacts Status: Production guidance

Overview

These checklists ensure Claude Code configuration follows Anthropic best practices for agents, skills, patterns, and CLAUDE.md files. Use during creation, review, and optimization.

When to use:

•Creating new agents or skills
•Reviewing configuration changes in PRs
•Auditing existing Claude Code setup
•Optimizing AI effectiveness

Agent Quality Checklist (7 Points)

1. YAML Frontmatter Complete and Valid ✅

• name: present (lowercase-hyphenated)
• description: present and action-oriented
• model: specified (sonnet or inherit)
• color: specified (optional but recommended)
• YAML syntax valid (no indentation errors)

Why: Frontmatter enables Claude Code to discover and invoke agents correctly.

Example:

yaml

---
name: dbt-expert
description: dbt transformation specialist. Consult proactively for model optimization, testing strategies, and performance issues.
model: sonnet
color: blue
---

2. Single-Purpose Focus ✅

• Agent has one clear responsibility
• Scope is neither too broad nor too narrow
• Role clearly defined in opening paragraph
• Expertise domain explicitly stated

Why: Single-purpose agents are easier to invoke correctly and provide more focused expertise.

Anti-pattern: "General data expert" (too broad) Good pattern: "dbt transformation specialist" (focused)

3. Detailed System Prompt with Examples ✅

• Opening paragraph states agent's role
• Core responsibilities listed (3-7 items)
• Expertise areas documented
• Example consultations provided
• Output format specified

Why: Detailed prompts improve agent response quality and consistency.

Sections to include:

•Role & Expertise
•Core Responsibilities
•Available Tools
•Example Consultations
•Quality Standards

4. Strategic Tool Restrictions ✅

• Only necessary tools granted
• Tool access rationale documented
• Restricted tools explicitly listed
• Security implications considered

Why: Limiting tools maintains focus and reduces security risk.

Example:

markdown

### ✅ Allowed Tools
- Read, Grep, Glob (analysis)
- WebFetch (documentation research)

### ❌ Restricted Tools
- Bash, Write (research-only agent)

5. Action-Oriented Description ✅

• Description uses active verbs
• Includes "Consult proactively" or similar
• Specifies when to delegate
• Triggers are clear

Why: Encourages Claude to delegate to specialists when appropriate.

Good: "Consult proactively for dbt optimization, testing, and performance issues" Bad: "Helps with dbt" (passive, vague)

6. Quality Standards Documented ✅

• Output format specified
• Quality requirements listed
• Validation protocols included
• Success criteria defined

Why: Ensures consistent, high-quality agent outputs.

Example:

markdown

## Quality Standards

**Every recommendation must include**:
- ✅ Analysis of current state
- ✅ Specific recommendations
- ✅ Implementation steps
- ✅ Validation approach

7. Integration with Agent Ecosystem ✅

• Related specialists documented
• Delegation patterns explained
• MCP tool usage specified (if applicable)
• Consultation protocol defined

Why: Agents work together effectively when integration is clear.

Skill Quality Checklist (5 Points)

1. Clear Workflow Steps ✅

• Steps numbered and sequential
• Each step has clear objective
• Tools to use specified
• Expected outputs defined
• Error handling included

Why: Step-by-step workflows ensure consistent execution.

Structure:

markdown

### 1. Step Name
**Execute**: [tools/commands to use]
**Output**: [expected result]
**Error handling**: [what to do if fails]

2. Error Handling and Rollback ✅

• Failure scenarios identified
• Graceful degradation defined
• Rollback procedures documented
• User feedback on errors

Why: Skills should handle failures gracefully and inform users.

Example:

markdown

## Error Handling

### Project Already Exists
**Check**: Directory exists
**Action**: Ask user to confirm overwrite or choose new name

### Git Branch Exists
**Check**: Branch exists
**Action**: Suggest alternative or checkout existing

3. Reusability Across Contexts ✅

• Works with different inputs
• No hardcoded project-specific values
• Templates use variables
• Tested with diverse scenarios

Why: Skills should be reusable, not one-off scripts.

Use variables: {project_name}, {date}, {user_input}

4. Quality Standards and Validation ✅

• Success criteria defined
• Output validation checks
• Quality requirements listed
• Acceptance tests specified

Why: Skills should validate their own outputs.

Example:

markdown

## Quality Standards

**Every project setup must**:
- ✅ Create all 4 core files
- ✅ Use consistent naming
- ✅ Create valid git branch
- ✅ Verify all files created

5. User-Facing Documentation ✅

• Purpose clearly stated
• Usage examples provided
• Trigger phrases listed
• Success metrics documented

Why: Users need to know when and how to use skills.

CLAUDE.md Quality Checklist (6 Points)

1. Specific Instructions, Not Generic Advice ✅

• Instructions are actionable
• Examples provided for key workflows
• Clear do's and don'ts
• No vague guidance

Why: Specific instructions improve Claude's effectiveness.

Bad: "Write good code" Good: "Use PEP 8 style guide. Run black . before committing."

2. Markdown Structure with Clear Headers ✅

• Logical hierarchy (H1 → H2 → H3)
• Table of contents if >300 lines
• Code blocks properly formatted
• Lists consistently styled

Why: Well-structured documentation is easier to parse and reference.

3. Import Patterns for Modularity (if >500 lines) ✅

• Main CLAUDE.md is focused (<300 lines ideal)
• Detailed patterns imported from .claude/rules/ and .claude/skills/reference-knowledge/
• Import syntax: @path/to/pattern.md
• Imported files are self-contained

Why: Modular structure improves readability and maintainability.

Example:

markdown

## Git Workflow
See @.claude/rules/git-workflow-patterns.md

## Testing
See @.claude/skills/reference-knowledge/testing-patterns/SKILL.md

4. Progressive Disclosure ✅

• Most important info first
• Quick reference at top
• Detailed sections below
• Advanced topics last

Why: Claude should find key information quickly.

Structure:

•Quick Start / Most Important
•Core Workflows
•Reference Documentation
•Advanced Topics

5. Version Control and Team Sharing ✅

• CLAUDE.md committed to version control
• .claude/agents/ committed
• Team-specific conventions documented
• Change history trackable

Why: Teams benefit from shared, versioned Claude Code configuration.

6. Regular Review and Updates ✅

• Last updated date included
• Obsolete sections removed
• New patterns incorporated
• Review schedule defined (quarterly recommended)

Why: Outdated instructions reduce AI effectiveness.

Pattern Quality Checklist (5 Points)

1. Production-Validated Solution ✅

• Used successfully in ≥2 real projects
• Results documented and verified
• Edge cases tested
• Confidence level assigned

Why: Only proven patterns should be documented.

Validation notes:

markdown

**Validated**:
- Project A: Reduced query time 80% (2024-09-15)
- Project B: Eliminated incremental issues (2024-10-03)
**Confidence**: 0.95 (High)

2. Decision Criteria ("When to Use") ✅

• Clear triggering conditions
• Alternative approaches mentioned
• Trade-offs documented
• Anti-patterns identified

Why: Helps decide when to apply the pattern.

Example:

markdown

## When to Use This Pattern

**Use when**:
- ✅ Incremental models with large datasets (>10M rows)
- ✅ Query runtime > 1 hour
- ✅ Source system supports incremental extraction

**Don't use when**:
- ❌ Small datasets (<1M rows)
- ❌ Full refresh is acceptably fast
- ❌ Source system lacks updated_at column

3. Real-World Examples ✅

• Concrete code examples
• Before/after comparisons
• Actual project references
• Screenshots or diagrams (if helpful)

Why: Examples make patterns immediately actionable.

4. Cross-References to Related Patterns ✅

• Links to related patterns
• Links to relevant agents
• Links to applicable skills
• External documentation referenced

Why: Patterns form a knowledge network.

Example:

markdown

## Related Patterns
- [dbt Testing Patterns](./dbt-testing-patterns.md)
- [Snowflake Optimization](./snowflake-optimization-patterns.md)

## Related Agents
- dbt-expert: Use this pattern for optimization consultations
- snowflake-expert: For warehouse-side performance tuning

5. Maintenance Information ✅

• Last validated date
• Next review date
• Known limitations
• Future improvements identified

Why: Patterns evolve as technologies and practices change.

Metadata:

markdown

**Last Validated**: 2025-10-15
**Next Review**: 2026-01-15 (quarterly)
**Confidence**: 0.92
**Limitations**: Requires dbt 1.6+, Snowflake warehouse

Usage Guide

During Creation

New Agent:

•Create agent file from template
•Use Agent Quality Checklist (7 points)
•Ensure all ✅ checked before committing

New Skill:

•Document workflow steps clearly
•Use Skill Quality Checklist (5 points)
•Test with 3+ diverse scenarios

New Pattern:

•Validate in ≥2 projects first
•Use Pattern Quality Checklist (5 points)
•Add confidence score

During PR Review

Review Checklist:

•Identify which checklists apply
•Verify all items checked
•Request changes for failures
•Approve when all items pass

During Optimization

Periodic Audit:

•Run all checklists against existing config
•Identify gaps or outdated content
•Update to meet current standards
•Document improvements made

Quality Metrics

Per-Artifact Quality Score

Formula: (Checked Items / Total Items) * 100

Thresholds:

•100%: Excellent - production ready
•85-99%: Good - minor improvements needed
•70-84%: Fair - significant gaps to address
•<70%: Poor - major rework required

Overall Configuration Health

Aggregate score across all:

•Agent files
•Skill files
•Pattern files
•CLAUDE.md

Target: ≥90% average across all artifacts

Continuous Improvement

After Every Project Completion

Review and Update:

• Did agents perform well? Update confidence levels
• Did skills work as expected? Refine workflows
• Did patterns apply? Update validation notes
• Is CLAUDE.md still accurate? Update instructions

Quarterly Configuration Audit

Systematic Review:

•Run all checklists on all files
•Calculate quality scores
•Identify lowest-scoring artifacts
•Prioritize improvements
•Update and re-validate

References

Anthropic Documentation

•Sub-agents: https://docs.claude.com/en/docs/claude-code/sub-agents.md
•Skills: https://docs.claude.com/en/docs/claude-code/skills.md
•Memory/CLAUDE.md: https://docs.claude.com/en/docs/claude-code/memory.md
•Hooks: https://docs.claude.com/en/docs/claude-code/hooks-guide.md
•Plugins: https://docs.claude.com/en/docs/claude-code/plugins.md

Internal Documentation

•Knowledge Organization: .claude/skills/reference-knowledge/knowledge-organization-strategy/SKILL.md
•Agent README: .claude/agents/README.md
•claude-code-expert: .claude/agents/specialists/claude-code-expert.md

Version: 1.0.0 Last Updated: 2025-10-21 Next Review: 2026-01-21 Source: Extracted from PR #164 (your-org/da-agent-hub) Maintainer: ADLC Platform Team