skills-rewrite

Public skill library for agentic coding, reasoning, review, and operations work.

This repo is intentionally flat:

• one top-level directory per skill
• one top-level README.md
• optional per-skill support material such as references/, examples/, or tests/

The goal is to keep every skill easy to browse, copy, remix, and publish without dragging around a larger framework project.

Included Skills

• couch-mode
• database-architecture
• infrastructure-deployment
• jetpack-developer
• lean-prover
• pal-commit
• p2p-engineer
• playwright-tester
• riff-onboarding
• riff-labs
• riff-obsidian
• storage-systems

Repo Rules

• Keep the repo public-safe.
• Prefer useful skills over clever ones.
• Do not import private operator modeling, internal company detail, or unstable lore.
• Do not import client, customer, or partner-specific knowledge into the core public skill set.
• Do not import pseudoscience, therapy material, or personality artifacts that are only meaningful in one private context.
• Rewrite skills as they land instead of dumping raw exports when a cleaner public version is easy to produce.
• Keep support files only when they materially improve the skill.
• Use Conventional Commits across Riff Labs repos.
• Pair that commit convention with Changesets wherever repo releases or versioned artifacts exist.
• Use Semantic Versioning for published versions.
• Prefer branches named TICKET-ID/ghusername/topic-name in Riff Labs repos so ownership and intent are visible immediately.

Layout

Each skill should look like this:

skill-name/
├── SKILL.md
├── references/        # optional
├── examples/          # optional
└── tests/             # optional

For imported skills that originally used different entrypoint names, normalize the public entrypoint to SKILL.md.

Current Direction

This first flattening pass focuses on practical engineering skills:

• low-input collaboration
• deployment and infra automation
• theorem proving
• commit hygiene
• website auditing
• storage architecture
• public Riff context

The next passes should expand carefully, skill by skill, with the same public-safety filter.

Install And Update

The repo ships with a simple top-level updater:

./update.sh

By default it will:

• upsert all top-level skills into ~/.codex/skills
• upsert all top-level skills into ~/.claude/skills
• write a version marker into each target root
• warn if the installed version is newer than the source you are trying to install

Use --symlink if you want to link the skill directories from this checkout instead of copying them.

Release Direction

The release model for this repo is:

• Conventional Commits for commit intent
• changesets for versioning and release notes
• Semantic Versioning for published versions
• CI/CD to cut tagged releases
• packaged skill artifacts for individual skills
• a tiny updater package for machines that should stay in sync automatically

The initial repository scaffolding for that model lives in:

• .changeset/
• .github/workflows/release.yml
• package.json
• skills-manifest.json

This repo is still in the first packaging pass. The target shape is:

• one release flow for the whole repo
• individual skill artifacts per release
• Debian packages for the updater and for installable skill bundles
• straightforward local installation into ~/.codex/skills and ~/.claude/skills