Shane Myrick's Skills Library

A personal library of reusable AI-agent workflows. Each skill has a portable SKILL.md core. OpenAI adapters supply interface metadata; the repository invocation policy also generates a Claude Code collection. The entire library ships as the smyrick-skills plugin, with portable Agent Plugins, Codex, and Claude Code packages.

Install

All skills as one plugin

Add this repository as a marketplace, then install the bundle:

Codex CLI (with plugin commands available):

codex plugin marketplace add smyrick/skills
codex plugin add smyrick-skills@shane-personal-plugins

Claude Code:

claude plugin marketplace add smyrick/skills
claude plugin install smyrick-skills@shane-personal-plugins

Each client reads its own catalog and receives the package with its invocation controls. The plugin includes all 11 skills and their resources. Choose the bundle or standalone copies for a given client to avoid duplicate skills. Existing installations are never automatically removed.

Portable, Codex, and Claude archives with checksums are attached to tagged GitHub releases. See plugin installation, version pinning, and development for details and current runtime validation coverage.

Individual skills / standalone collection

For Codex CLI, install every skill with the skills CLI:

npx skills add smyrick/skills --agent codex

Use npx skills add smyrick/skills --list to inspect the collection first. This installs the source collection, whose invocation controls are OpenAI-specific.

For Claude Code, clone this repository, build the target collection, then install that local collection:

npm ci
npm run setup:spec
npm run build:targets
npx skills add ./.dist/targets/claude --agent claude-code --copy

The generated SKILL.md files include Claude Code invocation controls. Use --copy to keep this variant in Claude's directory instead of overwriting the shared canonical skills directory used by other agents. Keep OpenAI and Claude target directories physically separate. Do not install the portable source directly into Claude Code if explicit-only invocation matters. These are Claude Code packages, not claude.ai upload packages. Other harnesses can read the portable instructions, but their invocation behavior is unverified.

Skill Index

Invocation is intentional. User skills are configured for explicit invocation in the OpenAI and generated Claude Code collections. Model skills may also be selected automatically or reached by another skill. Use the host's invocation syntax ($skill-name for standalone Codex skills, /skill-name for standalone Claude skills, or /smyrick-skills:skill-name for Claude plugin skills); live routing tests are separate from static validation.

The portfolio deliberately has no router today: the explicit skills are distinct, few enough to scan here, and their names map cleanly to deliberate workflows. Add a router only if observed invocation tests show that remembering the names is a recurring burden.

Use

Explicitly invoke a user skill by name:

Use $research-and-plan to research this change and produce an implementation plan.

research-orchestrator is the only model-invoked skill. Planning skills can reach it when durable, multi-pass research materially improves the result; a simple question should not trigger it.

Repository Contract

skills/
  <skill-name>/
    SKILL.md                 # Portable capability and workflow
    agents/
      openai.yaml            # OpenAI interface and invocation policy
    references/              # Optional details loaded only when needed
    assets/                  # Optional output templates or static assets

Portable frontmatter follows the Agent Skills specification, including its supported optional metadata. Runtime-specific fields belong in an adapter or generated target package. The repository currently excludes experimental tool approval declarations from the portable core.

Validate

Validation requires Node.js, Python 3.11+, and uv. Install the pinned upstream reference validator once before running checks:

npm ci
npm run setup:spec
npm run format
npm run build:plugins
npm run check
npm run oci:smoke

npm run check runs formatting checks, independently reported validation groups, and regression tests. Each group can also run directly:

The core check calls the upstream skills_ref.validate API; it does not maintain a second implementation of the spec. tools/skills-ref/pyproject.toml pins the source commit and uv.lock pins its dependencies. Checks run offline after setup and fail explicitly if the reference validator cannot run.

Formatting preserves YAML values and comments, including unknown fields; validation reports unsupported fields without deleting them. A malformed batch is rejected before any formatting writes.

npm run oci:smoke builds both target collections and verifies their actual archived contents, descriptors, paths, references, and index entries. Plugin checks require current committed packages and catalogs. npm run plugin:package builds three deterministic downloads; npm run plugin:verify independently checks their archived contents and checksums. Live client discovery, invocation, and execution report NOT RUN; schema and artifact parity checks do not prove runtime behavior. See validation coverage and source baselines.

OCI Package

Tagged releases publish the collection to GitHub Container Registry:

ghcr.io/smyrick/skills@sha256:<digest>
ghcr.io/smyrick/skills:vYYYY.MM.DD
ghcr.io/smyrick/skills:latest

Use digest pins when a consumer needs immutable installs. The archive retains the original skills/ collection and additionally includes targets/openai/skills/ and targets/claude/skills/. The index retains existing fields and adds per-target paths. The archive also includes the portable plugin.json, native plugins under plugins/, and both marketplace catalogs. Existing index fields and media types remain compatible.

Contributing

See CONTRIBUTING.md for invocation-mode decisions, authoring rules, tests, and the review checklist.

License

MIT — see LICENSE.