Agent Skill Kit

Agent Skill Kit (ASK)

Write an agent skill once. Deploy it to Claude Code, Codex, Gemini CLI, Cursor and Antigravity — in each tool’s native format.

Agent Skill Kit is a command-line package manager for AI agent skills. You define a skill once in a standard format; the ask CLI transforms it into the layout and file format each agent expects, then installs it globally or per-project. It also lints skills against per-agent token budgets and audits your library for skills that compete for the same prompts.

📖 Documentation & skill reference · Interactive command builder

brew tap NavanithanS/Agent-Skill-Kit && brew install agent-skill-kit
ask copy claude --skill ask-unit-test-generation

Agent Skill Kit Banner

Create once. Share across agents. See CHANGELOG.md for release notes.

🧠 Why Agent Skill Kit?

Managing instructions for multiple AI agents is tedious. You often have to:

ASK solves this by treating skills as reusable packages.

  1. Define Once: Write a skill in a standard format.
  2. Deploy Anywhere: ASK transforms and copies the skill to the correct location and format for each agent.
  3. Sync: Keep all your agents updated with a single command.

🚀 Features

📚 Skill library

All skills, with a link to each one’s reference page. Full index: browse the skill library.

Skill Category What it does
ask-bug-finder coding Best practices for systematic bug hunting and debugging
ask-code-reviewer coding An AI code reviewer that provides constructive feedback on code changes.
ask-commit-assistance coding Assist with code review and staging. NEVER automatically commits.
ask-component-scaffolder coding Standardizes UI component creation by generating a consistent folder structure and files.
ask-conceptual-integrity-sentinel coding Principal-level engineering agent that audits repositories for architectural drift, bloated abstractions, and “dead code.”
ask-db-migration-assistant coding Ensures safe database schema updates by requiring migration and rollback scripts before execution.
ask-docker-expert coding Expert guidance on Docker, Docker Compose, and container optimization. Focuses on multi-stage builds and security.
ask-effective-llm-coder coding Guides the agent in effective LLM-assisted coding using best practices for declarative workflows, simplicity, tenacity, and iterative refinement.
ask-explaining-code coding Explains code using analogies, ASCII diagrams, and step-by-step walkthroughs
ask-fastapi-architect coding Expert scaffolding for FastAPI projects. Enforces Pydantic V2, Async Database patterns, and Dependency Injection.
ask-flutter-architect coding Senior Flutter skill using FVM. Enforces project-specific standards: Provider, Layer-First Architecture, Stream-based Services, and strict coding conventions.
ask-flutter-mechanic coding Maintenance skill for Flutter projects using FVM. Handles clean builds, iOS/Android specific fixes, asset generation, and release protocols.
ask-impact-sentinel coding Guidelines for impact analysis, breaking change detection, strategic database design, and comprehensive database indexing review.
ask-laravel-architect coding Senior scaffolding skill. Handles SQL vs Mongo (Jenssegers/Official), SoftDeletes, and strict API standards.
ask-laravel-mechanic coding Senior maintenance skill. Enforces “Zero Data Loss” policies and handles Mongo/SQL debugging.
ask-nextjs-architect coding Expert scaffolding for Next.js 14+ (App Router). Enforces Server Components, Server Actions, and SEO best practices.
ask-owasp-security-review coding Conduct a thorough static security review of code, identifying vulnerabilities aligned with OWASP Top 10 risks, with severity ratings and remediation suggestions.
ask-python-refactor coding Best practices and guidelines for Python code refactoring
ask-readme-gardener coding Keeps documentation in sync with code by updating the README.md when features or APIs are added.
ask-refactoring-readability coding Refactor code to enhance readability, following principles like DRY, meaningful names, and modularization.
ask-security-sentinel coding Pre-flight security checker. Scans for exposed secrets and vulnerable patterns properly.
ask-shadcn-architect coding Strictly enforces shadcn/ui patterns, imports, and CLI usage when creating or modifying React UI components.
ask-shadcn-mechanic coding Expert maintenance skill for shadcn/ui. Handles component customization, responsive layout debugging, and Form/Zod wiring while strictly enforcing UI/UX design integrity.
ask-unit-test-generation coding Automates creation of comprehensive unit tests for functions or classes, emphasizing coverage of edge cases and assertions.
ask-vue-architect coding Expert scaffolding for Vue 3. Specialized for Laravel Inertia stacks, but supports Nuxt/Vite. Enforces Composition API & TypeScript.
ask-vue-mechanic coding Expert maintenance skill for Vue 3 within Laravel Inertia. Fixes navigation reloads, prop mismatches, and reactivity issues.
ask-adr-logger planning Automatically records Architectural Decision Records (ADRs) when a significant technical decision is made.
ask-brainstorm planning Explores user intent, requirements, and design before implementation. Required before any creative work.
ask-buildmaster planning Smart Epic Orchestration Agent - Acts as PM + Tech Lead + Delivery Manager for epic planning, execution, and delivery.
ask-hold-code planning Planning-mode lock that prevents code generation until the user explicitly approves implementation.
ask-project-memory planning Maintains a ‘Project Brain’ by recording architectural decisions and tech stack choices in a memory file.
ask-solution-architect planning Master Ideation and Strategic Architecture skill. Executes professional, multi-perspective ideation sessions utilizing SCAMPER, Six Hats, and Design Thinking.
ask-add-agent tooling How to add support for new AI code editors to Agent Skill Kit
ask-ast-mapper tooling Read-only subagent designed to generate lightweight AST (Abstract Syntax Tree) dependency maps.
ask-context-janitor tooling Aggressive token optimizer and context summarizer for AI orchestrators
ask-parallel-auditor tooling Orchestrator skill that splits a target repository into chunks and runs multiple audit subagents in parallel.
ask-pdf-processing tooling PDF text extraction, form filling, and merging using pypdf and pdfplumber
ask-skill-capture tooling Meta-skill. Analyzes the current session’s lessons and saves them as a permanent reusable skill.
ask-skill-creator tooling Teaches AI agents how to create skills for Agent Skill Kit
ask-smart-booking-test tooling An advanced autonomous testing skill for verifying end-to-end booking flows. Specializes in Flights, Movies, and Tours with integrated Payment Gateway testing.
ask-system-architect-prime tooling Principal Software Architect for repository audits, complexity analysis, and actionable refactoring recommendations
ask-wiki-init tooling Scaffolds the LLM Wiki pattern into any project — a persistent, compounding knowledge base maintained by AI agents.

📦 Installation

pip install agent-skill-kit

Homebrew (macOS / Linux)

brew tap NavanithanS/Agent-Skill-Kit
brew trust navanithans/agent-skill-kit
brew install agent-skill-kit

The brew trust step is required. Current Homebrew refuses to load formulae from third-party taps until you explicitly trust them, failing with “Refusing to load formula … from untrusted tap”. Review the formula before trusting it, as you should with any third-party tap.

If ask --version reports an unexpected version after installing, another copy is shadowing it on your PATH (a pip-installed one in an active virtualenv or conda environment is the usual cause). Check with type -a ask, and run /opt/homebrew/bin/ask --version to test the Homebrew install specifically.

From Source (Development)

# Clone the repository
git clone https://github.com/NavanithanS/Agent-Skill-Kit.git
cd Agent-Skill-Kit

# Install in editable mode
pip install -e .

Upgrading an Existing Installation

When a new version (like v0.4.0) is released, you must upgrade the CLI and then sync the new skills down to your local AI agents.

# 1. Upgrade the CLI
pip install --upgrade agent-skill-kit  # If using PyPI
brew upgrade agent-skill-kit           # If using Homebrew

# 2. Sync the new skills to your agents
ask update

🛠 Usage

0. Interactive Wizard ⭐

The easiest way to use ASK — let the wizard guide you.

ask wizard

Guides you through four actions: | # | Action | What it does | |—|——–|————-| | 1 | copy | Select skills + agent + scope → install | | 2 | purge | Select an agent → remove all ask-* skills | | 3 | sync | Sync every skill to every agent | | 4 | update | Update all installed skills to latest |

1. Copy Skills to an Agent

The primary way to use ASK — Deploy skills to your AI agents.

Interactive Mode (Recommended):

ask copy

The wizard guides you through:

  1. Stack detection — auto-suggests skills for your project (Laravel, Vue, Next.js, Flutter, FastAPI)
  2. Search/filter — type to narrow the skill list before selecting
  3. Skill selection — numbered table with descriptions and categories
  4. Agent selection — compatible agents highlighted for your chosen skill
  5. Destination — local (project) or global (user-wide)
  6. Conflict scan — all conflicts resolved upfront in one batch prompt

Quick Mode (with flags):

# Copy specific skill (fuzzy name matching — partial names work)
ask copy gemini --skill bug-finder
ask copy claude --skill fastapi-arch    # resolves → ask-fastapi-architect

# Copy all compatible skills to an agent
ask copy claude --all

2. List Available Skills

View your library of skills, including descriptions and supported agents.

# List all skills
ask list

# Search for skills
ask list --search "docker"

# Filter by category or agent
ask list --category coding --agent claude

3. Manage Rules (ask rules compile)

Easily manage rule files (.cursorrules, CLAUDE.md, .agent/rules/) by compiling them from a single central Universal Truth directory!

# Place your `.md` snippets in .agents/rules/ or ~/.agents/rules/
ask rules compile

4. Interactive Purge (ask purge)

Safely delete ask-* skills from selected target agents (or the entire Universal directory) with interactive selection prompts.

ask purge

5. Validate Library

Check your skill library for errors, missing metadata, or circular dependencies.

ask validate

6. Create a New Skill

AI-Assisted (Recommended): Simply ask your AI agent to create a skill for you:

Create a new skill for Docker best practices

Make a skill that teaches REST API design

create new skill: explaining-code
Purpose: Explains code using analogies, ASCII diagrams, and step-by-step walkthroughs. Triggered by queries like "How does this work?"
Instructions emphasize conversational tone, multiple analogies, and highlighting common misconceptions.

Prerequisites: Deploy ask-skill-creator to your agent first (see Tooling Skills).

Manual CLI (Alternative): Launch the interactive wizard to generate a standardized skill template.

ask create skill

7. Sync All Skills

Synchronize your entire skill library to all supported agents at once.

ask sync all

8. Update Skills

Keep your installed skills up-to-date with the latest versions from the repository.

ask update

Features:

9. Add Support for New Agents

Want to use Windsurf or Aider? Use the scaffold wizard:

ask add-agent

This creates the necessary adapter code, making the new agent available instantly.

10. Skill Development Tools (New in v0.2.0)

# Lint skills for token limits and schema compliance
ask skill lint

# View token usage report
ask skill profile

# Generate manifest.json for routing
ask skill compile

11. Evaluate Skills (ask test)

Validation checks that a skill is well-formed; ask test checks that skills don’t fight each other for the same prompts — the silent killer of a growing library.

# Offline trigger/collision audit (no API key, CI-friendly)
ask test

# Audit a single skill
ask test ask-laravel-architect

# Fail the build on collisions/misses (opt-in CI gate)
ask test --strict

# Machine-readable output
ask test --json

Each skill declares paraphrased user prompts in tests/evals.yaml:

should_fire:
  - "set up a new laravel api with mongodb models"

The audit ranks each prompt against the whole library (TF-IDF) and flags collisions — similar skills competing for the same prompt. It’s a fast lexical pre-screen, not a model of how an agent semantically routes; live-model behavioral evals (ask test --behavior) are planned.

12. Run as an MCP Server (ask mcp)

Expose your skill library to any MCP-capable agent so it can discover and pull skills at runtime — no pre-deployment required. Read-only by design (it never writes to your filesystem).

# Start the server (stdio)
ask mcp serve

# Inspect the exposed tools without starting a server
ask mcp tools

# Dry-run a search to preview what an agent would see
ask mcp probe "scaffold a fastapi service with async db"

Tools exposed: list_skills, search_skills, get_skill. Example client config:

{
  "mcpServers": {
    "agent-skill-kit": { "command": "ask", "args": ["mcp", "serve"] }
  }
}

Install the optional dependency with pip install "agent-skill-kit[mcp]".

🎯 Supported Agents

Agent Local Path (Project) Global Path (User) Format
Antigravity .agent/skills/ ~/.gemini/antigravity/skills/ SKILL.md (YAML)
Gemini CLI .gemini/skills/ ~/.gemini/skills/ SKILL.md (YAML)
Claude Code .claude/commands/ ~/.claude/commands/ Markdown Command
Codex codex.md ~/.codex/instructions/ Markdown
Cursor .cursor/rules/ ~/.cursor/rules/ Markdown Rules

� Available Skills

ASK comes with a curated collection of skills to boost your AI agent’s capabilities. Each skill provides specialized instructions and best practices.

Planning Skills

Skill Description Use Cases
adr-logger Automates creation of Architectural Decision Records • Recording tech decisions
• Documenting context & consequences
• Maintaining decision history
hold-code Enforces a strict planning-mode lock • Reviewing design documents
• Waiting for explicit user approval
• Preventing premature code generation
brainstorm Guidelines for exploring user intent and requirements • Defining user intent
• Gathering requirements
• Exploring design options
project-memory Maintains a ‘Project Brain’ of decisions • Avoiding re-discussions
• Checking past decisions
• Recording new choices
buildmaster Smart Epic Orchestration Agent • Epic discovery & scoping
• Ticket decomposition
• Execution tracking & handoff
solution-architect Master Ideation and Strategic Architecture • Multi-perspective ideation
• Designing robust systems
• Structured output via frameworks

Coding Skills

Skill Description Use Cases
bug-finder Best practices for systematic bug hunting and debugging • Debugging complex issues
• Isolating bugs
• Using debugging tools
code-reviewer AI code reviewer providing constructive feedback • Code quality checks
• Security & performance review
• Learning best practices
effective-llm-coder Guides agent in declarative, simple, tenacious coding • Declarative workflows
• Simplicity & tenacity
• Iterative refinement
commit-assistance Assist with code review, staging, and committing • Pre-commit review
• Meaningful commit messages
• Staging files
explaining-code Explains code using analogies and diagrams • Understanding complex code
• Visualizing flow
• Learning new codebases
flutter-architect Senior Flutter skill using FVM • Layer-First Architecture
• Stream-based Services
• Strict coding conventions
flutter-mechanic Maintenance skill for Flutter projects using FVM • Clean Build Protocol
• iOS/Android fixes
• Release protocols
laravel-architect Senior scaffolding skill for Laravel (SQL/Mongo) • Logic Layer separation
• Hybrid SQL/Mongo Relations
• Test-Driven Scaffolding
laravel-mechanic Senior maintenance skill for database safety • Zero Data Loss protocol
• N+1 Query prevention
• Queue debugging & forensics
owasp-security-review Static code analysis aligned with OWASP Top 10 • Security scanning
• Identifying vulnerabilities
• Compliance checks
python-refactor Guidelines for Python code refactoring • Improving code quality
• Refactoring legacy code
• Python best practices
refactoring-readability Improves code structure for clarity • Renaming vars/functions
• Reducing complexity
• Improving readability
impact-sentinel Guidelines for impact analysis and database design • Breaking change detection
• Strategic DB design
• Safe optimization
unit-test-generation Automates creation of comprehensive unit tests • Generating new tests
• Covering edge cases
• Improving coverage
vue-architect Expert scaffolding for Vue 3 (Inertia/Nuxt) • Component blueprints
• Stack detection
• Best practices enforcement
vue-mechanic Expert maintenance skill for Vue 3 (Inertia) • Fixing navigation reloads
• Debugging prop mismatches
• Solving reactivity issues
component-scaffolder Standardizes UI component creation • Consistent folder structure
• Typed props
• Auto-generating tests
db-migration-assistant Ensures safe database schema updates • Drafting migrations
• Creating rollback scripts
• Preventing data loss
readme-gardener Keeps documentation in sync with code • Updating API docs
• Documenting new features
• Maintaining README accuracy
shadcn-architect Enforces shadcn/ui patterns and consistency • Preventing style bloat
• Enforcing import rules
• Promoting accessibility
shadcn-mechanic Expert maintenance for shadcn/ui components • Customizing variants (CVA)
• Debugging form validation
• Fixing Tailwind layout limits
security-sentinel Pre-flight security checker for secrets/vulns • Blocking commits with secrets
• Detecting SQL injection
• Flagging unsafe Blade usage
nextjs-architect Expert scaffolding for Next.js 14+ (App Router) • Server Components
• Server Actions
• Metadata API
fastapi-architect Expert scaffolding for FastAPI projects • Pydantic V2
• Async SQLAlchemy
• Dep Injection
docker-expert Expert guidance on Docker & Containers • Multi-stage builds
• Security best practices
• Image optimization
conceptual-integrity-sentinel Audits repositories for architectural drift and bloat • Detecting complexity bloat
• Identifying dead code
• Enforcing simplicity

Tooling Skills (Meta-Skills)

📚 ask-wiki-init

Description: Scaffolds the LLM Wiki pattern into any project — a persistent, compounding knowledge base maintained by AI agents.

How to Use:

ask copy antigravity --skill ask-wiki-init

Use Cases:

🧹 ask-context-janitor

Description: Aggressive token optimizer and context summarizer for reducing large text files, logs, or agent outputs.

How to Use:

ask copy antigravity --skill ask-context-janitor

Use Cases:

�️ ask-ast-mapper

Description: Read-only subagent for generating lightweight AST dependency maps and structural overviews of directories.

How to Use:

ask copy antigravity --skill ask-ast-mapper

Use Cases:

⚡ ask-parallel-auditor

Description: Orchestrator skill that splits a target repository into chunks and runs multiple audit subagents in parallel to bypass context limits.

How to Use:

ask copy antigravity --skill ask-parallel-auditor

Use Cases:

�🛠️ ask-skill-creator

Description: Teaches AI agents how to create skills for Agent Skill Kit

How to Use:

# Deploy to all agents so they can create skills
ask sync all

Use Cases:

Example Workflow:

  1. Deploy this skill to your agent: ask copy gemini --skill ask-skill-creator
  2. Ask your agent: “Create a skill called ‘ask-docker-best-practices’ for containerization guidelines”
  3. The agent generates the skill files automatically

🎯 ask-add-agent

Description: How to add support for new AI code editors to Agent Skill Kit

How to Use:

# Deploy to help your agent add new editor support
ask copy antigravity --skill ask-add-agent

Use Cases:

Example Workflow:

  1. Deploy this skill to your agent
  2. Run the wizard: ask add-agent
  3. Or ask your agent to help: “Add support for Windsurf editor”
  4. The agent follows the documented process to create the adapter

📄 ask-pdf-processing

Description: Handle PDF text extraction, form filling, and merging

How to Use:

ask copy antigravity --skill ask-pdf-processing

Use Cases:

🧠 ask-skill-capture

Description: Meta-skill. Analyzes the current session’s lessons and saves them as a permanent reusable skill.

How to Use:

# Deploy to your agent (e.g., Antigravity, Gemini)
ask copy antigravity --skill ask-skill-capture

Use Cases:

Example Workflow:

  1. You struggle through a task and finally get it right
  2. You say: “Capture this as a skill called ‘ask-deployment-protocol’”
  3. The agent analyzes the conversation and generates the SKILL.md
  4. You verify and save it

🏗️ ask-system-architect-prime

Description: Principal Software Architect for repository audits, complexity analysis, and actionable refactoring recommendations

How to Use:

ask copy antigravity --skill ask-system-architect-prime

Use Cases:

Example Workflow:

  1. Deploy this skill to your agent
  2. Ask: “Audit this repository for architectural flaws”
  3. The agent generates an ARCHITECTURAL_AUDIT.md with health score, burn list, and recommendations

🚀 Quick Start with Skills

# Guided wizard (new in v0.6.0)
ask wizard

# Or use the docs site Command Builder — no terminal needed
# → https://navanithans.github.io/Agent-Skill-Kit/docs/

# View all available skills
ask list

# Deploy a specific skill to an agent
ask copy gemini --skill ask-bug-finder --global

# Deploy all compatible skills to an agent
ask copy claude --all

# Sync all skills to all agents
ask sync all

# Create your own skill (interactive)
ask create skill

�📐 Skill Format

Each skill is a directory containing:

[!IMPORTANT] Naming Convention: All skill names must start with the ask- prefix (e.g., ask-bug-finder, ask-commit-assistance).

# skill.yaml
name: ask-bug-finder
version: 1.0.0
category: coding
agents:
  - gemini
  - claude
  - cursor
depends_on:
  - ask-code-reviewer  # Automatically installed when bug-finder is installed

SKILL.md Format (v2.0)

As of v0.2.0, SKILL.md files use a token-optimized format:

---
name: ask-example
description: Brief description
triggers: ["trigger phrase 1", "trigger phrase 2"]
---

<critical_constraints>
❌ NO forbidden actions
✅ MUST required actions
</critical_constraints>

<heuristics>
- condition → action
</heuristics>

🧩 Design Principles

  1. Universal Definition: Skills are defined in a neutral format that can be adapted to any agent.
  2. Local-First, Global-Ready: Prioritize project-specific skills (checked into git) while supporting user-wide global skills.
  3. Safe by Default: The CLI will never silently overwrite an existing skill. It always asks.
  4. Agentic Workflow: The toolkit includes skills (skill-creator) specifically designed to help AI agents help you build more skills.

🗂 Repository Structure

agent-skill-kit/
├── ask/                     # CLI Source Code
│   ├── commands/            # copy, sync, purge, wizard, create, …
│   └── utils/               # adapter loader, filesystem helpers
├── agents/                  # Adapters for each AI agent
│   ├── gemini/
│   ├── claude/
│   ├── codex/
│   ├── antigravity/
│   └── cursor/
├── skills/                  # The Skill Library (38 skills)
│   ├── coding/
│   ├── planning/
│   ├── tooling/
│   └── manifest.json        # Auto-generated by ask skill compile
├── wiki/                    # Persistent LLM-maintained Knowledge Base
│   ├── index.md             # Top-level content catalog
│   ├── log.md               # Chronological activity log
│   └── SCHEMA.md            # Structural conventions
├── docs/                    # GitHub Pages site
│   └── index.html           # Generated by scripts/generate_site.py
└── scripts/
    └── generate_site.py     # Static site generator

🤝 Contributing

Contributions are welcome!

  1. Create a Skill: Use ask create skill and submit a PR with your best prompts.
  2. Add an Agent: Use ask add-agent, test it, and submit the new adapter.

License

MIT

This site is open source. Improve this page.