Claude Toolkit

A portable, reusable Claude Code toolkit — a centralized library of skills, rules, hooks, scripts, and templates that can be symlinked into any project to establish consistent agentic programming practices.

Why

1. Eliminate duplication — Skills like /handoff, /run-tests, and /update-docs are shared across projects. One place to improve them.
2. Token efficiency — Wrap deterministic operations in CLI scripts so the LLM orchestrates rather than manually executing tool calls.
3. Knowledge ingestion — Structured way to feed web content into the toolkit via /research, analyze it, and propose improvements — all gated by user approval.

Skills

Skill	Description
/assess-skills-coverage	Audits skills library for workflow gaps
/condense-repo	Identifies duplicate code, unused files, redundant patterns
/update-docs	Syncs Claude context documents after work
/handoff	Session handoff — syncs docs, summarizes work, prepares next session
/run-tests	Auto-detects and runs test suite (pytest, jest, vitest, go test, cargo test)
/research	Fetches web content, analyzes it, proposes toolkit improvements
/plan	Explores codebase and designs implementation plans
/review	Reviews code changes for quality and rule compliance
/setup-toolkit	Sets up this toolkit in another repository via symlinks

Scripts

Script	Description
scripts/fetch_url.py	Fetch URL and convert HTML to markdown
scripts/file_inventory.py	Directory tree summary with file metadata
scripts/diff_summary.py	Structured git diff parsing
scripts/setup_toolkit.py	Symlink/copy toolkit into target repos

All scripts use Python stdlib only — no pip dependencies required.

Rules

Behavioral guardrails auto-loaded by Claude Code:

• coding-standards — Clean naming, single-responsibility, explicit error handling
• test-integrity — Never modify tests to match broken code
• session-hygiene — Run /update-docs and /handoff at session end
• approval-gate — All /research proposals require user approval
• token-efficiency — Prefer scripts over multi-tool-use chains

Setup

In this repo (development)

Clone and work directly. Skills, rules, and scripts are all here.

In another repo (consuming)

# From the target repo:
python /path/to/claude-toolkit/scripts/setup_toolkit.py --target .

# Or use the skill:
/setup-toolkit /path/to/target-repo

This will:

• Symlink skills from the toolkit into your .claude/skills/
• Copy rules to your .claude/rules/ (for per-project customization)
• Merge hooks into your .claude/settings.json
• Copy CLAUDE.md template if you don't have one

Use --copy instead of symlinks if your environment doesn't support them.

Repository Structure

claude-toolkit/
├── .claude/
│   ├── settings.json          # Hooks and permissions
│   ├── PLANNING.md            # Development roadmap
│   ├── TASKS.md               # Session log
│   ├── skills/                # The core product (9 skills)
│   └── rules/                 # Auto-loaded behavioral rules (5 rules)
├── scripts/                   # CLI utilities (Python, stdlib-only)
├── templates/                 # Scaffolds for consuming repos
├── proposals/                 # Staging area for /research output
└── outputs/                   # Generated reports

Design Decisions

• Symlinks over submodules — Instant propagation, single source of truth
• Stdlib-only scripts — No dependency management needed
• Rules copied, skills symlinked — Rules customize per-project; skills stay identical
• Proposal staging — /research writes to proposals/ and stops; nothing auto-applied
• Multi-language detection — Skills auto-detect project language rather than hardcoding one framework