Specialized agent skills that streamline common developer workflows and reduce token usage.
Each skill is a self-contained directory with a SKILL.md, Node.js implementation scripts, and a test suite. Claude Code installs natively via the plugin marketplace; other agent harnesses can vendor any skill directory via npx.
Register this repository as a marketplace, then install individual skills through the /plugin UI:
/plugin marketplace add castastrophe/agent-skills
/plugin install gh-notification-summary@agent-skillsClaude Code clones the repo, mounts each skill directory, and surfaces them under /plugin for enable/disable.
Use the npx installer to vendor a skill directory into any agent's skill folder. The installed payload is a plain directory with SKILL.md + scripts — no Claude-specific wiring.
# List available skills
npx @allons-y/agent-skills
# Install a skill to the default location (~/.claude/skills/)
npx @allons-y/agent-skills gh-notification-summary
# Install to your agent's skill directory
npx @allons-y/agent-skills gh-notification-summary --dir ~/.config/cursor/skills
# Install all skills
npx @allons-y/agent-skills --allFor Node-runtime skills with dependencies, run npm install in the installed directory once it's vendored.
Note:
.zip-based distribution is deprecated and will be removed in the next major version. The installer now copies the skill directory directly.
| Skill | Description | Trigger |
|---|---|---|
gh-notification-summary |
Review, summarize, and manage GitHub notifications via an interactive local dashboard | "check my GitHub notifications", /unsub <number>, "mark all done" |
The package exposes a getSkills() helper for tool builders who want to list or load skills dynamically:
import { getSkills } from "@allons-y/agent-skills";
const skills = getSkills();
// [
// {
// name: 'gh-notification-summary',
// path: '/path/to/skills/gh-notification-summary',
// description: 'Review, summarize, and manage GitHub notifications...',
// mdPath: '/path/to/skills/gh-notification-summary/SKILL.md'
// }
// ]- Node.js (v24), supports
nvm use - Yarn 4 (via Corepack)
yarn installThis is a Yarn workspaces monorepo — each skill under skills/ is its own workspace. Target a specific skill with yarn workspace:
yarn workspace @allons-y/skill-gh-notification-summary test
yarn workspace @allons-y/skill-gh-notification-summary lintOr run across all skills at once:
yarn workspaces foreach -A run testFor full setup instructions — running tests, linting, evals, and publishing — see CONTRIBUTING.md.
agent-skills/ # Root workspace (publishes to npm)
├── package.json # workspaces: ["skills/*"]
├── index.js # Exports getSkills()
├── bin/install.js # npx installer CLI
├── .claude-plugin/
│ ├── marketplace.json # Auto-generated Claude Code marketplace
│ └── plugin.json # Auto-generated plugin manifest
├── scripts/ # Root orchestration scripts
│ ├── run-tests.js # Parallel Node test runner
│ └── generate-plugin-manifest.js # Generates .claude-plugin/*.json
├── skills/ # Yarn workspace members
│ └── <skill-name>/ # Each skill is a workspace
│ ├── package.json # private: true; declares `skill.runtime`
│ ├── SKILL.md # Metadata and usage docs
│ ├── scripts/ # Node.js implementation
│ ├── tests/ # `node --test` suite
│ └── evals/ # Eval prompts (evals.json)
└── .github/
└── workflows/ # CI and release automationHow do I install a skill for use with Claude Code?
/plugin marketplace add castastrophe/agent-skills
/plugin install gh-notification-summary@agent-skillsClaude Code clones the repo, reads .claude-plugin/marketplace.json, and mounts each skill directory.
Can I use these skills with agents other than Claude Code?
Yes. Each skill is a plain directory with a SKILL.md (frontmatter: name, description) and a scripts/ implementation. Use the npx installer to vendor a skill into any agent harness's skill folder:
npx @allons-y/agent-skills gh-notification-summary --dir <your-agent-skills-dir>Or call getSkills() from the package to enumerate skills programmatically.
How do I add a new skill or run tests?
See CONTRIBUTING.md for the full guide: environment setup, running tests, linting, evals format, and PR checklist.
Contributions are welcome — new skills, improvements to existing ones, bug fixes, and documentation. See CONTRIBUTING.md for guidelines.
Ideas for new skills:
- GitHub PR review summarizer
- Linear / Jira issue triage
- Slack digest summarizer
- Daily standup generator from git log
Apache 2.0 — use freely, modify as needed, contribute back if you can.
