A collection of Claude Code skills and tools for enhanced development workflows.

ai-debate-hub/
├── skills/ # Claude Code skills
│ ├── debate/ # Three-way debate skill
│ └── validation/ # Comprehensive app validation skill
├── tools/ # Reusable tools and frameworks
│ ├── retrospective-learning.md
│ ├── RETROSPECTIVE-LEARNING-INTEGRATION-GUIDE.md
│ └── sample/ # Reference implementations
│ ├── feature-validation-SKILL.md
│ └── report-template.html
├── debates/ # Generated debate transcripts
└── tests/ # Test files

Location: skills/debate/ Purpose: Three-way debates between Claude, Gemini CLI, and OpenAI Codex CLI

Claude is both a participant and moderator, contributing its own analysis alongside other AI advisors. Enables multi-round discussions where all three systems analyze problems independently and converge on recommendations through genuine debate.

Key Features:

• Multi-round debates (1-10 rounds configurable)
• Session persistence across rounds
• Multiple debate styles: quick, thorough, adversarial, collaborative
• Automatic synthesis of perspectives
• Web viewer for debate visualization

Usage:

/debate Should we use Redis or in-memory cache for our session store?
/debate -r 3 -d thorough Review our authentication implementation

Full Documentation →

Location: skills/validation/ Purpose: CLI-agnostic browser validation with evidence-based reporting.

Discovers routes, elements, and critical flows, then validates them in a real browser with screenshots, redacted snapshots, console/error checks, failed-network grouping, observed outcomes, resumable state, and an HTML report.

Backend strategy:

• Default: agent-browser
• Fallbacks/adapters: project-local Playwright, playwright-cli, Playwright MCP, Claude/Codex Chrome MCP tools, and browser-harness

Usage:

/validate http://localhost:3000
/validate --interactive
/validate --backend agent-browser --mode standard http://localhost:5173
/validate --backend playwright-local --config validation.config.json --mode standard http://localhost:5173
/validate --fresh
/validate --resume

For deterministic runs, skills/validation/scripts/select-backend.js can recommend the available backend from the target project before execution.

Output: test-manifest/validation-state.json, evidence files under test-manifest/evidence/, and HTML reports under test-manifest/reports/. Project-local Playwright runs may use validation.config.json for route expectations, permission-aware access-denied handling, breakpoints, and extra redaction values.

Location: skills/audit/ Purpose: Multi-skill technical due diligence pipeline for codebase audits.

The bundle includes the audit orchestrator, setup method, hard-stop checks, Tambon hunt, blind-spot walk, 13 domain audits, audit decision handling, fix-prompt generation, and audit-loop iteration.

Install all audit skills by copying the children of skills/audit/ into your active skill root:

cp -r skills/audit/* ~/.codex/skills/
cp -r skills/audit/* ~/.claude/skills/

Primary entry points:

• audit-orchestrator: full /audit style due diligence.
• audit-loop: audit, roadmap, remediation, rerun, and skill patching loop.
• audit-fix-generator: produce remediation prompts for individual findings.

Location: skills/project-setup/, skills/project-curate/, skills/project-doctor/, skills/handoff/, skills/skillify/

These are reusable workflow skills:

• project-setup: generate or refresh CLAUDE.md and AGENTS.md from a real codebase.
• project-curate: tune a project's .claude/ directory with targeted rules/skills/agents.
• project-doctor: read-only audit of project orchestration scaffolding.
• handoff: write a structured session handoff for clean context transfer.
• skillify: turn a successful repeated workflow into a reusable skill.

Location: skills/mercadopago-integration/ Purpose: MercadoPago OAuth/payment integration guidance for Supabase/React-style apps.

The repo copy was checked against the installed skill copies; content differences were line-ending-only for the compared files, so the repo version remains the canonical copy.

Location: tools/sample/feature-validation-SKILL.md Purpose: Reference implementation showing retrospective learning integration

This is an example skill demonstrating how retrospective learning has been integrated into a real-world validation skill. It includes:

• Phase 1: Trigger capture (reads session transcript)
• Phases 2-9: Feature testing workflow
• Phase 10: Retrospective (evidence-based self-evaluation)

Use this as a reference when integrating retrospective learning into your own skills.

Report Template: tools/sample/report-template.html

Location: tools/retrospective-learning.md Purpose: Enables skills to learn from execution through evidence-based retrospectives

A reusable module that adds self-evaluation and learning capabilities to any skill by:

1. Capturing trigger context - Documents why the skill was invoked
2. Tracking execution - Records all actions during execution
3. Performing retrospectives - Evaluates performance against original intent with session transcript evidence

Key Benefits:

• Skills learn from mistakes through line-cited evidence
• User corrections are documented as lessons
• Skill improvements suggested based on actual execution
• Works with large session transcripts (> 256KB handled correctly)

Features:

• Cross-platform (Windows/macOS/Linux)
• Size-aware transcript reading
• Evidence-based self-assessment
• Skill improvement recommendations

Full Documentation →

Location: tools/RETROSPECTIVE-LEARNING-INTEGRATION-GUIDE.md Purpose: Step-by-step guide for adding retrospective learning to your skills

A practical guide showing how to integrate the retrospective learning module into existing skills with minimal changes.

What it provides:

• Copy-paste prompt template for Claude
• Real before/after example (353 → 528 lines, +49%)
• Section-by-section breakdown of what gets added
• Verification checklist to ensure correct integration
• Troubleshooting for common failures (rewrites, missing content)
• Actual diff showing feature-validation integration

Quick Start:

Backup my SKILL.md with timestamp, then apply retrospective-learning.md to it.
CRITICAL:
- Create backup: SKILL.md.backup-{timestamp}
- Modify SKILL.md directly
- DO NOT rewrite existing content
- ONLY add Phase 1 trigger capture + final phase retrospective
- Renumber phases correctly
- Line count should increase ~40-60% (not double)
Use feature-validation-SKILL.md as reference.

[Full Integration Guide →](tools/RETROSPECTIVE LEARNING-INTEGRATION-GUIDE.md)

git clone https://github.com/wolverin0/ai-debate-hub.git
cd ai-debate-hub

Copy skills to your Claude Code skills directory:

# Debate skill
cp -r skills/debate ~/.claude/skills/
# Validation skill
cp -r skills/validation ~/.claude/skills/
# Audit bundle
cp -r skills/audit/* ~/.claude/skills/
# Workflow skills
cp -r skills/handoff skills/skillify skills/project-setup skills/project-curate skills/project-doctor ~/.claude/skills/
# Or link instead of copy
ln -s $(pwd)/skills/debate ~/.claude/skills/debate
ln -s $(pwd)/skills/validation ~/.claude/skills/validation

Debate skill:

# Basic usage
/debate Should we implement feature X?
# With options
/debate -r 3 -d thorough Review our API design

Validation skill:

# Basic usage
/validate http://localhost:3000
# With options
/validate --resume # Resume from previous validation
/validate --fresh # Start fresh

Follow the [Retrospective Learning Integration Guide](tools/RETROSPECTIVE LEARNING-INTEGRATION-GUIDE.md) to add self-evaluation capabilities to your own skills.

The debate skill generates a web-based viewer for visualizing debates:

cd debates
python -m http.server 8000
# Open http://localhost:8000/viewer.html

Viewer features:

• Side-by-side round comparison (all 3 participants)
• Synthesis view (final recommendations)
• Full chronological transcript
• State/metadata debug view

• Claude Code CLI
• Gemini CLI - npm install -g @google/gemini-cli
• OpenAI Codex CLI - npm install -g openai-codex

• Running web application (dev server or static site)
• Recommended backend: agent-browser
• Optional backends: playwright-cli, Playwright MCP, Claude/Codex Chrome MCP tools, or browser-harness

• Claude Code (already required for skills)
• Session transcripts must be accessible at ~/.claude/projects/{project-slug}/*.jsonl

• Debate Skill Documentation - Full debate skill usage guide
• Validation Skill - Adapter-based browser validation skill
• Feature Validation Example - Reference implementation with retrospective learning

• Retrospective Learning Module - Technical documentation for the retrospective learning module
• Integration Guide - How to add retrospective learning to your skills

Before integration:

## Phase 1: Analyze Implementation
git diff --name-only HEAD~5
## Phase 2: Plan Test Scenarios
...
## Phase 9: Generate HTML Report

After integration (+175 lines, +49%):

## 🚨 CRITICAL BLOCKERS
You MUST read session transcript before and after execution
## Phase 1: Analyze Implementation
**🚨 MANDATORY: Identify Trigger Context from Session Transcript**
[80 lines of trigger capture instructions]
git diff --name-only HEAD~5
## Phase 2-9: [Original content, renumbered]
...
## Phase 10: Retrospective & Learning
**🚨 MANDATORY: Read Session Transcript for Evidence-Based Retrospective**
[80 lines of self-evaluation instructions with line citations]

Result: Skill now captures why it was invoked, tracks actions, and performs evidence-based retrospectives citing actual transcript line numbers.

Contributions welcome! Areas of interest:

• New debate styles or moderator modes
• Additional skills with retrospective learning integration
• Improvements to retrospective learning module
• Better visualization in debate viewer
• Bug fixes and documentation improvements

To contribute:

1. Fork the repository
2. Create a feature branch
3. Make your changes
4. Submit a pull request

MIT License - See LICENSE file for details

• Repository: github.com/wolverin0/ai-debate-hub
• Issues: Report bugs or request features
• Claude Code: anthropics/claude-code

• v1.0.0 - Initial release
  • Debate skill with three-way participation
  • Auto-healing module and integration guide
  • Feature validation reference implementation
  • Web-based debate viewer

Built with Claude Code Demonstrating self-improving skills through retrospective learning and collaborative AI debate.