Claude Code Skills

A collection of reusable workflow patterns (skills) for Claude Code that enable structured planning and autonomous development.

Overview

These skills provide structured workflows for:

• Autonomous Development - Headless task execution with Research → Implement → Verify cycles
• Adversarial Review - Post-completion QA that explores features as a user, writes failing tests, creates fix tickets
• Project Planning - Comprehensive feature planning through research and interview cycles
• Test-Driven Development - Red-green-refactor workflow for all implementation

Installation

git clone https://github.com/wedow/skills.git ~/.claude/skills
cd ~/.claude/skills
./setup.sh

This creates:

• ~/CLAUDE.md - Symlink to PRIME DIRECTIVE (maximal simplicity policy)
• ~/.local/bin/ai - Autonomous development loop runner

Also install the ticket system: See https://github.com/wedow/ticket for installation instructions.

Ensure ~/.local/bin is in your PATH:

export PATH="$HOME/.local/bin:$PATH"

Usage and Philosophy

The autonomous development loop is inspired by Geoffrey Huntley's Ralph Wiggum technique - put an agent in a while loop and let it iterate toward completion. The core insight is one agent session equals one task. AI tools are massively effective when focused. Results nosedive as context accumulates and distracts. Tools like tk enable multi-session handoff and coordination - agents create tasks capturing outstanding work and relevant context, then exit. The next agent starts fresh with a clean slate.

Typical Workflow

Start a fresh Claude Code session for each skill invocation:

1. project-vision - Capture project philosophy and goals through structured interview
2. project-planning - Generate comprehensive feature plan (close session after)
3. reviewing-plans - Refine the plan (run 1-3 times in fresh sessions until ready)
4. plan-to-tickets - Convert finalized plan into self-contained tickets
5. ai - Run the autonomous development loop until complete or blocked

The loop automatically invokes autonomous-development, investigate-blocker, and adversarial-review as needed, stopping only when human input is required or all tasks are verified complete.

Skills Reference

Core Development Skills

Skill	Description
autonomous-development	Picks up tasks from tickets and executes Research → Implement → Verify cycles autonomously
adversarial-review	Post-completion QA: explores features as a user, writes failing tests for real issues, creates fix tickets
test-driven-development	Enforces red-green-refactor: write failing test first, then minimal code to pass

Planning Skills

Skill	Description
project-vision	Captures project philosophy through structured interview
project-planning	Generates comprehensive feature plans through research and user interviews
reviewing-plans	Refines plans for implementation readiness, classifies issues as blocking/non-blocking
plan-to-tickets	Converts finalized plans into self-contained tickets for autonomous execution

Investigation & Resolution

Skill	Description
investigate-blocker	Deep parallel research to resolve REQUIRES-INVESTIGATION tasks

Meta Skills

Skill	Description
create-new-skill	Guide for creating new skills with proper structure and best practices

Typical Workflow

1. project-vision     → Capture project goals and philosophy
2. project-planning   → Generate detailed feature plan
3. reviewing-plans    → Refine until implementation-ready
4. plan-to-tickets    → Create tickets for autonomous execution
5. ai                 → Execute tasks, then adversarial review before declaring done

Running Autonomous Development

Single invocation:

claude -p "Use autonomous-development skill"

Continuous loop (runs until all tasks complete or human input needed):

ai

Signal controls while the loop is running:

• Ctrl+C — kill the active claude instance and exit immediately
• Ctrl+\ — let the current claude instance finish, then exit after the iteration completes

Ticket System

Skills use tk for local task management. Install from: https://github.com/wedow/ticket

Always run tk help first to confirm command syntax.

Tickets are stored in .tickets/ as markdown files with YAML frontmatter.

Skill Structure

Each skill is a directory containing SKILL.md with YAML frontmatter:

---
name: skill-name
description: Brief description of what it does and when to use it
---

# Skill Content
...

Skills follow a coordinator/sub-agent pattern:

• Main agent coordinates workflow and dispatches sub-agents
• Sub-agents perform actual research, implementation, and verification
• Sub-agents should read the project's CLAUDE.md as their first action

License

MIT