Skip to content

Repository files navigation

nitays-agent-toolkit

A curated, best-of-breed AI coding toolkit that combines top community skills with battle-tested workflows. Works as a git submodule for any project, instantly supercharging your AI-assisted development.

Built for Claude Code.

What's Inside

Category Count Source
Skills 103 obra/superpowers, anthropics/skills, mobile-agent-toolkit, custom
Agents 20 architect, code-reviewer, verifier, and more
Commands 14 create-pr, fix-issue, evolve, and more
Hooks 20+ Quality, security, observability, ECC
Core rules 7 Prose standards in rules/*.md (referenced from CLAUDE.md)

Quick Start

Option A: Per-project setup (recommended if you also use work repos)

Cleanest separation -- your personal tools stay in personal projects, zero chance of mixing with org repos.

The shell scripts were replaced by a single Go CLI, acr. Build it once:

# One-time: clone to your home directory and build the CLI
git clone https://github.com/nitayk/nitays-agent-toolkit.git ~/nitays-agent-toolkit
cd ~/nitays-agent-toolkit && go build -o ~/.local/bin/acr ./cli   # ensure ~/.local/bin is on PATH

See cli/README.md for full command docs.

# Link the checkout into a project and deploy skills/agents/commands/hooks
acr link ~/projects/my-app
cd ~/projects/my-app && acr sync          # add --copy to copy instead of symlink

# Restart Claude Code -- done! 103 skills deployed into that project's .claude/.

To update a project later:

cd ~/nitays-agent-toolkit && git pull && acr update   # refresh community sources
cd ~/projects/my-app && acr sync                 # re-deploy

Option B: Git submodule (team/open-source projects)

Version-pinned assets committed to the project repo. acr install adds the submodule, syncs, and installs a post-merge hook that re-syncs after git pull:

git submodule add https://github.com/nitayk/nitays-agent-toolkit.git .cursor/rules/shared
go build -o ~/.local/bin/acr ./.cursor/rules/shared/cli
acr install

The checkout lives at .cursor/rules/shared on purpose: Claude Code auto-loads everything under .claude/, so a submodule there would blow up the context window. Claude ignores .cursor/, making it a safe home for the checkout.

What goes where

acr sync deploys into the project's .claude/ directory:

What Location How
Skills .claude/skills/ Copied (Claude Code doesn't index symlinked skill trees)
Agents .claude/agents/ Symlinked by default (--copy to copy)
Commands .claude/commands/ Symlinked by default (--copy to copy)
Hooks .claude/hooks/ + .claude/hooks.json Merged

Skills Inventory

Planning & Design

Skill Description
brainstorming Socratic design refinement before coding
writing-plans Break work into 2-5 min bite-sized tasks
executing-plans Batch execution with human checkpoints
prd-generation PRD generation via structured conversation
task-breakdown Break PRDs into parent tasks and sub-tasks

Development Workflows

Skill Description
tdd-workflow Red-Green-Refactor cycle
test-driven-development RED-GREEN-REFACTOR with anti-patterns reference
git-workflow Stable-state commit workflow for complex tasks
pr-workflow Full PR lifecycle (pre-checks, monitoring, merge)
create-pr Quick PR creation with template detection
fix-issue Fix a GitHub issue end-to-end
setup-local-dev Persistent dev server with pm2
finishing-a-development-branch Merge/PR/cleanup decision workflow

Code Quality

Skill Description
best-practices-enforcement Validates code against universal standards
code-cleanup Three-phase cleanup (AI slop, dead code, anti-patterns)
code-review-excellence Effective code review practices
code-optimization Performance and efficiency optimization
code-migration Code migration patterns
requesting-code-review Pre-review checklist
receiving-code-review Responding to feedback with rigor
verification-before-completion Evidence before claims
security-audit Security vulnerability scanning

Debugging & Testing

Skill Description
systematic-debugging 4-phase root cause process
debug-workflow Log-Reproduce-Fix cycle
webapp-testing Playwright-based web app testing
test-until-pass Iterative test-fix loop

Multi-Agent & Parallel

Skill Description
dispatching-parallel-agents Concurrent subagent workflows
subagent-driven-development Fresh subagent per task with two-stage review
multi-agent-branching Feature-branch isolation for concurrent agents
using-git-worktrees Parallel development branches
mass-repo-orchestration Multi-repo batch operations
council Multi-perspective decision making

Research & Knowledge

Skill Description
deep-research Multi-source research with citations
search-first Research before coding
onboard-developer Developer onboarding guide
generate-docs Auto-generate documentation
generate-changelog Generate changelogs from git history
doc-coauthoring Collaborative document creation
address-pr-feedback Systematic PR feedback handling

Frontend & Documents

Skill Description
frontend-design Production-grade frontend interfaces
web-artifacts-builder Multi-component HTML artifacts
agent-browser Browser automation skill
docx Word document creation, editing, tracked changes
pdf PDF extraction, creation, merge/split, forms
xlsx Spreadsheets with formulas, formatting, analysis
pptx PowerPoint creation and editing
gdoc Google Docs integration

Meta & Tools

Skill Description
skill-creator Guide for creating effective skills
writing-skills Create new skills following best practices
mcp-builder Guide for building MCP servers
session-memory Persistent context across sessions
using-superpowers Introduction to the skills system
agent-token-optimization LLM token cost optimization
continuous-learning-v2 Instinct-based learning system
strategic-compact Context compaction at logical intervals
prompt-optimizer Optimize LLM prompts
repository-organization Restructure messy repos

Optional Power-Ups

These are full frameworks best used via their own install mechanisms:

  • BMAD-METHOD (36k stars): npx bmad-method install - Full agile AI development with 12+ specialized agents
  • wshobson/agents (28k stars): /plugin marketplace add wshobson/agents - 112 agents, 146 skills, 73 plugins
  • obra/superpowers marketplace: /plugin marketplace add obra/superpowers-marketplace - Plugin system for Claude Code

MCP Configuration

See mcp/example-config.json for a ready-to-use configuration with public MCP servers:

  • context7 - Library documentation lookup
  • playwright - Browser automation
  • puppeteer - Browser testing
  • github - GitHub API integration
  • filesystem, git, shell - Local development tools

Adding More Skills

Skills are just markdown files. You can add more from:

Install with: npx skills add owner/repo --skill "skill-name"

Updating

Pull latest + update community sources

cd ~/nitays-agent-toolkit
git pull
acr update                        # fetch latest from obra/superpowers + anthropics/skills
git add -A && git commit -m "chore: update community skills"

Re-apply to projects

# Linked or submodule project
cd ~/projects/my-app && acr sync

See SOURCES.md for full provenance tracking.

Credits

Built on the shoulders of giants:

License

MIT License - see LICENSE for details.

Skills from external sources retain their original licenses:

  • obra/superpowers: MIT License
  • anthropics/skills: Apache 2.0 (example skills), source-available (document skills)

About

Curated AI coding toolkit: 34 skills + 200+ rules for Cursor & Claude Code. Best of obra/superpowers, anthropics/skills, and custom workflows.

Resources

Stars

1 star

Watchers

0 watching

Forks

Releases

Packages

Contributors

Languages