Agent Skills that make a repo shippable, from "works on my machine" to something a stranger can trust, run, and that you can release. The build-side counterpart to distro-skills: ship it well, then distribute it.
Works with Claude Code, OpenAI Codex, Cursor, Windsurf, and any agent that supports the Agent Skills spec.
ship-context runs first and writes .agents/ship.md (stack, package manager, test tooling, publish target) that every other skill reads. ship-checklist audits the repo against a shippable bar and sequences the rest.
Foundation
- ship-context: detect the stack and record it in
.agents/ship.md; every skill reads it. - ship-checklist: audit the repo against a shippable bar and sequence the rest.
Quality
- add-tests: scaffold the right test runner and write first meaningful tests.
- set-up-ci: run tests and lint on every push and PR, with a green badge.
- lint-and-format: add the standard formatter and linter, applied and enforced.
- security-basics: a disclosure policy, dependency and secret scanning, no committed secrets.
Package and release
- package-for-npm: a clean, npx-runnable npm package.
- package-for-pypi: a clean, pipx-runnable PyPI package.
- cut-a-release: semver bump, tag, release notes, publish.
- write-a-changelog: a Keep a Changelog file kept current.
Repo health
- repo-hygiene-pass: LICENSE, CONTRIBUTING, code of conduct, templates, gitignore.
- readme-that-converts: a README first screen that says what, why, and how in ten seconds.
- license-and-legal: pick and add the right license and notices.
- env-and-config: document config, keep secrets out, fail loudly on missing vars.
- one-command-run: make it runnable in one command and verify the stranger path.
Copy the skill folders you want from skills/ into your agent's skills directory (for Claude Code, ~/.claude/skills/), or point your agent at this repo. Each skill is a self-contained folder with a SKILL.md.