Imported from galligan/skills (AGENTS.md). Install upstream with npx skills add galligan/skills. Copyright stays with the author.

galligan/skills

Author shareable skills with Skillset. Distribute complete individual skills from skills/.

Paths below are relative to the repository root.

Source ownership

• Edit repository guidance in .skillset/partials/repository.md. Skillset compiles it to AGENTS.md and .claude/CLAUDE.md; the files in .skillset/rules/ and .skillset/_claude/ select the output surfaces.
• Edit skill source under .skillset/skills/<name>/ and workspace configuration in skillset.yaml.
• Edit reusable authoring references under .skillset/shared/references/. If an installed skill needs a shared reference, declare that reference in the skill’s resources so Skillset includes it in the generated bundle.
• Do not hand-edit the generated skills/ tree, AGENTS.md, .claude/CLAUDE.md, or either skillset.lock. Change the source and rebuild.
• Keep the single public output root configured by codex.skills.path: skills. Do not add plugin bundles or parallel provider skill output roots. Claude compilation supplies project instructions only.
• If an installed skill needs a file, include that file in the generated skill bundle.
• Document each skill’s required external tools and services.
• An installed skill must work without the authoring checkout or undeclared plugin services.
• Keep credentials and private machine paths out of public skill content. Keep Python bytecode and virtual environments out of source skill directories because the compiler copies sibling files.

Authoring and communication

• Before adding or reviewing a persistent directive, apply .skillset/shared/references/instruction-selection.md to decide whether the directive earns its place.
• When choosing or changing an instruction’s owning surface, use .skillset/shared/references/instruction-placement.md.
• When writing skills, agent definitions, prompts, or other agent-facing instructions, use .skillset/skills/agentish/SKILL.md. It routes to the shared Agentish language guide.
• This repository selects People Words as its communication style. For human-facing replies and prose, apply .skillset/skills/people-words/SKILL.md. Preserve the intended audience and document form; Agentish's controlled instruction conventions do not apply to ordinary prose.
• For a requested systematic instruction audit, use .skillset/skills/redliner/SKILL.md.

If a change alters skill activation, required or prohibited behavior, instruction ownership, or provider routing, review the changed canonical source and affected consumers with Redliner before handoff. Record findings, corrections, and remaining limits. For wording-only edits, check preserved meaning and affected links; do not run a repository-wide audit solely for a typo.

Use the pinned runtime and compiler versions. Follow README.md for build and change-record commands and docs/publishing.md for validation and distribution. Record version changes through Skillset’s change and release commands. Commit source, change evidence, and generated output together.

Use a branch and review the diff before committing. Publishing, installation, and activation are separate from building.