skill

ai-toolkit

Personal AI toolkit for Claude Code, Codex, and OpenCode: skills, commands, configs and tools

irega/ai-toolkitcommunitylistedunreviewed

Set up

Skills are installed by placing their folder where your agent loads skills.

Download ZIP from GitHub Open the repository

The download comes from github.com/irega/ai-toolkit itself, not from our server, so what you get is exactly what its author published.

Install instructions come from the README

Skills do not use an MCP config. Follow the steps in the README below.

README

synced Oct 25, 2026

ai-toolkit

Personal AI toolkit for Claude Code, Codex, and OpenCode: skills, commands, configs and tools. Clone it on a new machine, run the installer, start working.

Requirements

macOS with Homebrew
npm (get it via nvm)

Install

git clone https://github.com/irega/ai-toolkit.git
cd ai-toolkit
./scripts/install.sh

It also symlinks everything in skills/ into ~/.claude/skills, ~/.codex/skills, and ~/.config/opencode/skills, and registers MCP servers into whichever of Claude Code/Codex/OpenCode are installed.

What it installs

Tools

Tool	Purpose
RTK	CLI proxy that cuts token usage on dev commands
OpenSpec	Spec-driven planning, run `openspec init` per project to enable
CodeGraph	Local code knowledge graph MCP server, run `codegraph init` per project to enable
Caveman	Ultra-compressed communication mode, cuts token usage ~75% (requires Node)
GitHub CLI	`gh` — used by Claude Code for PRs, issues, checks, releases
Engram	Persistent memory MCP server (SQLite + FTS5), agent-agnostic
CodeBurn	Local AI coding token and cost tracking across tools and agents

Skills

Skill	Purpose
handoff	Compact the current conversation into a handoff doc for another agent
delivering-changes	Full delivery run: prepare, plan, implement, verify, review, and open draft PRs, in order, with gates between phases
preparing-projects	Detect spec conventions, index with CodeGraph, check RTK, recover Engram memory
planning-changes	Plan a change: spec/plan with acceptance criteria, risks, tests, tasks
implementing-tasks	Implement a plan: test-first per task, RED/GREEN report
verifying-changes	Verify a change against its spec: repo checks, evidence per criterion, E2E, spec drift
reviewing-changes	Review a diff from fresh context, one verdict per lens, report only
opening-pull-requests	Open draft PRs: one per independent concern, repo template, English
Humanizer	Rewrite AI-sounding prose to read naturally, installed cross-agent via `npx skills add`
Ponytail	Forces minimal, lazy-first code solutions, installed cross-agent via `npx skills add`
Superpowers	Process skills (brainstorming, TDD, systematic debugging, plan writing, code review...), installed cross-agent via `npx skills add`

MCP servers

Defined once in configs/mcp-servers.json (tool-agnostic), registered into each installed CLI natively (eg: claude mcp add).

Server	Purpose
playwright	Browser automation (Chrome extension mode)
Azure DevOps	Backlog, work items and pull requests (optional, one instance per configured organization)

Some servers may need env vars (e.g. an extension token). Copy .env.example to .env.local and fill it in before running scripts/mcp/sync-mcp.sh — values get injected into the server registration.

Scripts

install.sh runs all of these, but each can also be run standalone:

Script	Purpose
`scripts/link-skills.sh`	(Re-)symlink `skills/` into `~/.claude/skills`, `~/.codex/skills`, `~/.config/opencode/skills`
`scripts/unlink-skills.sh`	Remove broken skill symlinks from all three skill directories
`scripts/mcp/sync-mcp.sh`	Dispatcher: registers MCP servers from `configs/mcp-servers.json` into whichever of claude/codex/opencode are installed
`scripts/mcp/sync-mcp-<assistant>.sh`	Register MCP servers with one specific assistant (`claude`, `codex`, or `opencode`) only
`scripts/agents/sync-agents.sh`	Dispatcher: generates one pinned subagent per (tier, effort) pair in `skills/delivering-changes/tiers.json` for whichever of claude/codex/opencode are installed (details: `scripts/agents/README.md`)
`scripts/agents/sync-agents-<assistant>.sh`	Generate pinned subagents for one specific assistant (`claude`, `codex`, or `opencode`) only, rendered from `scripts/agents/templates/`
`scripts/agents/prune-agents.sh`	Remove `delivery-*` pinned agents that current `tiers.json` would not generate — leftovers from a renamed/removed tier or an old naming scheme