eins78/agent-skills

Collection of Agent Skills for Claude Code and compatible AI coding agents.

Looking for Plot? The git-native planning workflow has moved to its own repo: eins78/plot

Skills

Skill	Description
ai-review	Get AI code review from a second model (Gemini/OpenAI) mid-session via CLI
apple-mail	Read email via Apple Mail.app and AppleScript (read-only)
apple-notes	Read Apple Notes via AppleScript (read-only)
ballot	Durable decision artefact for async decisions — reviewed over chat, on a PR, on a train, after the session ends. Must/Should/Could tiers, empty checkboxes, reconciliation in sessionlog. Single async decider or multi-reviewer panel; pairs with dossier or stands alone (ADRs, PR review, hiring, vendor, household)
bye	Session wrap-up — reconstructs history, creates sessionlog, commits, summarizes next steps
chrome-browser	Dedicated Chrome with CDP for Playwright MCP — persistent sessions, launchd-managed, Cloudflare tips
dossier	Structured research — ranked recommendations, cited sources, parallel subagents, optional decision ballots
lab-notes	Experiment management — hypothesis, append-only running log, verdict. Rigorous + Lite modes.
pandoc	Document format conversion — 60+ formats via pandoc instead of ad-hoc scripts
tmux-control	Reliable tmux patterns — targeting, send-keys, capture-pane, wait-for sync, monitoring
tracer-bullets	Thin vertical slice before widening — reduce uncertainty by building end-to-end first
typescript-strict-patterns	TypeScript patterns — tsconfig, ESLint strict, Zod, discriminated unions, branded types

Installation

As a Claude Code / Cursor plugin (recommended — auto-updates)

Register the marketplace and install:

/plugin marketplace add eins78/agent-skills
/plugin install eins78-skills@eins78-marketplace

Skills auto-update when you run /plugin update.

Via skills CLI

pnpx skills add https://github.com/eins78/agent-skills.git --global --agent claude-code --all --yes

Single skill (via plugin marketplace)

Individual skills are listed in the marketplace and can be installed by name:

/plugin install bye@eins78-marketplace

Manual (single skill)

# Symlink a single skill
ln -s ~/CODE/agent-skills/skills/typescript-strict-patterns ~/.claude/skills/typescript-strict-patterns

# Or copy it
cp -r ~/CODE/agent-skills/skills/typescript-strict-patterns ~/.claude/skills/

Skills are picked up automatically by Claude Code based on their globs frontmatter.

Format

Each skill follows the Agent Skills spec — a SKILL.md file with YAML frontmatter (name, description, globs) and markdown instructions.

Creating Skills

Quick Start

my-skill/
├── SKILL.md          # Required: frontmatter + instructions
├── README.md         # Required: development notes, design decisions, testing history
└── REFERENCE.md      # Optional: detailed reference material (>100 lines)

SKILL.md Template

---
name: my-skill-name     # lowercase, hyphens, max 64 chars, must match directory
description: What it does and when to use it. Include trigger keywords. (max 1024 chars)
# Optional fields per agentskills.io spec:
compatibility: claude-code, cursor
license: MIT
metadata:
  source: https://github.com/your-org/your-repo
  version: "1.0.0"
---

# My Skill Name

[Instructions here — keep under 500 lines]

See [REFERENCE.md](REFERENCE.md) for details.

Note: Reference files over 100 lines should include a table of contents.

Key Principles

1. Be concise — Only add what Claude doesn't already know
2. Progressive disclosure — Overview in SKILL.md, details in referenced files
3. Third person — "Processes files" not "I help you process files"
4. One level deep — Reference files directly from SKILL.md, avoid nesting
5. Use checklists — Multi-step workflows benefit from copy-paste checklists
6. Test across models — Haiku may need more guidance than Opus

Skill Composition

Skills can reference other skills by name in their instructions:

## Committing Your Fix

When ready to commit, use the **commit-notation** skill:
- Bugfixes use `B` intention
- Risk level depends on test coverage

Claude loads both skills when context matches. No formal import system — just mention by name.

Official Resources

Specification & Docs:

• Agent Skills Specification — Open standard format
• Best Practices — Anthropic's authoring guide
• Claude Code Skills — Claude Code integration

Example Skills & Tools:

• Anthropic Skills Repository — Official example skills
• skill-creator — Scaffolding tool for new skills

Development

Adding a changeset

When making changes, add a changeset to describe your change:

pnpm changeset

Edit the created file to set the bump type (patch, minor, or major), describe the change, and add a <!-- bumps: --> block naming the affected skills:

<!--
bumps:
  skills:
    skill-name: minor
-->

Releasing

Releases are automated via GitHub Actions. When a PR with changesets merges to main:

1. The Action creates a "Version Packages" PR with version bumps and CHANGELOG updates
2. Merging the version PR creates git tags and a GitHub Release

Manual release (fallback):

GITHUB_TOKEN=$(gh auth token) pnpm run version   # bumps skill + plugin versions, writes changelog, syncs metadata
pnpm run release   # runs version, commits, tags — then push with: git push --follow-tags

Validation

pnpm test          # skills parse correctly
pnpm run validate  # SKILL.md frontmatter checks (name, version, license)

License

MIT