AgentSkillKit by Catalayer

Universal workflow skills for Claude Code, Codex, Qwen Code, DeepSeek, Kimi, Cursor, and AI coding agents.

Most coding agents can edit files. Few know how to ship safely.

AgentSkillKit gives agents reusable engineering workflows for release audits, bug-fix loops, README polish, npm package checks, Chrome extension QA, AI code review, and security scans. It is not a prompt dump. It is a practical workflow layer for agent-assisted software development.

Why skills instead of prompts?

A raw prompt tells an agent what you want. A skill tells an agent how to do it safely and consistently.

Every skill in AgentSkillKit defines:

• Role — what the agent is acting as
• Scope — exactly which files it may touch (and which it may not)
• Required workflow — ordered, verifiable steps
• Safety rules — what the agent must never do
• Verification checklist — how the agent confirms it finished correctly
• Final report format — structured output, not a prose essay

The result is predictable, auditable agent behavior instead of one-shot guesses.

Supported agents

AgentSkillKit skills are plain Markdown with YAML frontmatter. They work with any agent that accepts text instructions.

| Agent | How to use | Guide | |---|---|---| | Claude Code | Paste rendered skill into terminal session | adapters/claude | | Codex | Pass skill as system or task context | adapters/codex | | Qwen Code | Paste into prompt or AGENTS.md | adapters/qwen | | DeepSeek / Deep Code | Paste rendered skill into session | adapters/deepseek | | Kimi Code | Paste rendered skill into Kimi workspace | adapters/kimi | | Cursor | Paste into Composer or .cursor/rules | adapters/cursor | | AGENTS.md agents | Drop skill body into project AGENTS.md | adapters/generic-agents-md |

AgentSkillKit vs RepoBlackbox

| Tool | Purpose | When | |---|---|---| | AgentSkillKit | Gives agents reusable, scoped engineering workflows | Before the agent acts | | RepoBlackbox | Snapshots and audits what the agent actually changed | After the agent acts |

AgentSkillKit tells agents what to do before they act.
RepoBlackbox verifies what changed before and after they act.

They are complementary. See docs/relation-to-repoblackbox.md.

Quick Start

AgentSkillKit is currently install-from-source only. npm publishing will follow after stabilization.

git clone https://github.com/stephenywilson/AgentSkillKit
cd AgentSkillKit
npm install
npm run build
npm link
agent-skill-kit --help

Commands

agent-skill-kit init

Initialize AgentSkillKit in your current project.

agent-skill-kit init             # creates config + example local skill
agent-skill-kit init --force     # overwrite existing config and example

Creates .agent-skill-kit/ with config.json, skills/, workflows/, rules/, and a starter skills/example-custom-skill.md.

agent-skill-kit list

List all available skills (built-in and local).

agent-skill-kit list

agent-skill-kit show <skill>

Show skill metadata, safety rules, checks, and a body preview.

agent-skill-kit show repo-public-release-audit
agent-skill-kit show bug-fix-loop

agent-skill-kit use <skill>

Render a skill as a copy-paste-ready prompt. Use --var to inject project-specific values.

agent-skill-kit use github-release-polish \
  --var project_path=/path/to/repo \
  --var repo_url=https://github.com/user/repo \
  --var version=1.0.0

The rendered body goes to stdout. Metadata goes to stderr — so you can pipe:

agent-skill-kit use security-secret-scan \
  --var project_path=/path/to/repo \
  > prompt.txt

agent-skill-kit validate

Validate all built-in and local skill files.

agent-skill-kit validate

Checks: required frontmatter, body quality, variable declarations, no private paths, no secrets, no prohibited chain-of-thought instructions.

Built-in Skills

SKILL.md format (v0.2+)

New skills use a directory-based SKILL.md format with structured sections including workflow steps, verification checklists, and a Chinese summary.

| Skill | Category | Purpose | |---|---|---| | repo-public-release-audit | release | Full audit before making a private repo public | | bug-fix-loop | debugging | Structured bug-fix cycle: reproduce → isolate → fix → verify | | npm-package-release-check | release | Pre-publish npm checklist: build, fields, dry-run, privacy | | github-readme-polish | documentation | Polish a README for public GitHub quality | | security-secret-scan | security | Scan for secrets, private paths, and .env leaks | | ai-code-review | review | Structured code review with scope boundaries | | chrome-extension-qa | testing | Pre-submission QA for Chrome extensions | | x-launch-post | marketing | Draft a concise, accurate X post for a project launch |

Legacy flat-file format (v0.1)

The original 12 skills remain fully functional:

| ID | Category | Purpose | |---|---|---| | readme-audit | documentation | Audit README quality | | github-release-polish | release | Final release checklist | | security-privacy-scan | security | Privacy and secrets scan | | ui-screenshot-audit | design | UI/UX screenshot review | | agent-safe-refactor | refactoring | Scoped, behavior-preserving refactor | | repo-url-fix | maintenance | Fix clone URLs and install commands | | changelog-update | release | Update CHANGELOG and version fields | | npm-package-release-check | release | npm publish verification | | python-package-release-check | release | PyPI publish verification | | cli-smoke-test | testing | Generate or improve CLI smoke tests | | open-source-launch | release | Full open-source launch checklist | | docs-structure-review | documentation | Review docs structure and gaps |

Local Custom Skills

Add your own skills under .agent-skill-kit/skills/ in any project.

cd /path/to/your-project
agent-skill-kit init
agent-skill-kit list
agent-skill-kit show example-custom-skill

Local skills appear in list with a local source label, and override built-in skills with the same id (a warning is printed).

See docs/custom-skills.md for the full guide.

Skill Format

Two formats are supported:

Legacy flat-file (v0.1): skills/<name>.md — Markdown with YAML frontmatter and a body prompt.

SKILL.md directory format (v0.2): skills/<name>/SKILL.md — richer format with structured sections.

See docs/skill-format.md and docs/writing-skills.md.

Examples

Project-level AGENTS.md templates for common project types:

• JavaScript / npm project
• Python CLI project
• Chrome extension project
• Open-source launch

Safety Principles

Every built-in skill follows these rules:

• Scope boundaries — the agent is told exactly which files it may touch
• No auto-publish — publishing commands are always printed, never executed
• No feature additions — skills do one specific thing
• Observable reasoning — agents report Plan, Files inspected, Changes made, Commands run, Results, and Remaining risks
• Privacy-first — validate checks all skills for private paths, secrets, and prohibited instructions

Writing Your Own Skills

1. Run agent-skill-kit init — creates .agent-skill-kit/skills/example-custom-skill.md
2. Copy and edit the example
3. Run agent-skill-kit validate to check it
4. Run agent-skill-kit use my-skill --var project_path=/path/to/project

See docs/writing-skills.md for the full guide.

Roadmap

• [x] 12 legacy flat-file skills
• [x] Local custom skill loading
• [x] SKILL.md directory format
• [x] 7 agent adapter guides
• [ ] agent-skill-kit new <skill> — interactive scaffolder
• [ ] Workflow chains: run multiple skills in sequence
• [ ] npm package release

License

Apache-2.0 — see LICENSE.

© 2024-2026 Catalayer AI