Skills

Generic Agent Skills source repo and tooling.

This repository stores skills in the open Agent Skills format used by npx skills and compatible agents. Each skill lives in its own directory under skills/ and contains a required SKILL.md file with YAML frontmatter.

Quick Start

Install Node dependencies:

npm install

List the skills discoverable from this repo:

npm run skills:list

Validate all skills and print size/token budget estimates:

npm run skills:validate

Skills

• attribution — configurable attribution rules for GitHub, Linear, Azure DevOps, and similar systems.
• brain — persistent cross-session memory workflow backed by an Obsidian vault.
• deep-dive — exhaustive project onboarding workflow that writes durable notes into the brain vault.
• skill-creator — Anthropic's Apache-2.0 skill for creating, reviewing, evaluating, and improving Agent Skills.

Third-Party Skills

• skills/skill-creator/ is copied from anthropics/skills and keeps its bundled LICENSE.txt.

Install

List the skills in this repo:

npm run skills:list

Install every skill globally for every supported agent:

npx skills add . --global --all --full-depth

Install a single skill globally:

npx skills add . --global --skill brain --full-depth

Install into the current project instead of globally by omitting --global:

npx skills add . --skill attribution --full-depth

Configure Attribution

The attribution skill does not hardcode a personal name or emoji. Generate the local config before installing or syncing the skill:

npm run configure:attribution -- --name "My Agent" --emoji "<emoji>"

This writes skills/attribution/references/attribution-config.md, which is ignored by git.

Configure Brain

The brain skill does not hardcode a vault name or path. Generate the local config before installing or syncing the skill:

npm run configure:brain -- --vault-name "my-vault" --vault-path "/absolute/path/to/vault"

This writes skills/brain/references/brain-config.md, which is ignored by git. Without it, the skill stops on first use and asks for configuration.

Development

Type-check TypeScript scripts:

npm run typecheck

Validate one skill:

npm run skills:validate:attribution

The validator is skills/skill-creator/scripts/quick_validate.py. It checks Agent Skills frontmatter and reports static budgets:

• Description character count and estimated tokens.
• SKILL.md line count and estimated tokens.
• Body line count and estimated tokens.
• Warnings for descriptions over 1024 characters, bodies over 500 lines, or bodies over roughly 5000 estimated tokens.

Token counts are estimates using a chars/4 heuristic. They are useful for keeping skills lean, not exact model accounting.

Layout

skills/
  <skill-name>/
    SKILL.md
    scripts/      # optional
    references/   # optional
    assets/       # optional

The name field in each SKILL.md must match its parent directory name.

Repo Guide

Use this repo as the source of truth for skills. Installed global OpenCode or Claude skills should be treated as deployed copies, not edited directly.

When adding a skill:

1. Create skills/<skill-name>/SKILL.md.
2. Keep the name frontmatter equal to <skill-name>.
3. Put trigger guidance in description.
4. Keep core instructions in SKILL.md; move long docs to references/.
5. Put deterministic helper code in scripts/.
6. Run npm run skills:validate and npm run skills:list.

When editing a vendored skill:

• Keep its license file with the skill.
• Note local changes in this README if they matter.
• Prefer small patches over rewriting upstream content.

Generated local config files, like attribution identity, should stay ignored by git.