title: Glossary description: Definitions of the terms Repertoire uses: skill, catalog, catalog trust, loose catalog, target, scope, manifest, lock file, managed copy, source-qualified ID, provenance, and dry run.

Glossary¶

If you have used a Linux package manager, the model is familiar: catalogs are repositories, add installs a package, and repertoire.yaml is the manifest you commit so others get the same set. Terms defined here are underlined throughout the documentation; hover one for the short definition.

Skill¶

A directory containing a SKILL.md with name and description frontmatter, followed by instructions for an AI coding agent. It may include supporting files such as templates or scripts.

The format is shared by many agents, which is what makes skills portable: the same directory works in Codex, Claude Code, Cursor, Gemini CLI, and the rest. Each agent simply looks for it in a different place.

Catalog¶

A Git repository (or a local folder) that publishes skills through a repertoire.yaml with a catalog: section.

The built-in phillarmonic catalog is preconfigured and holds Phillarmonic's official skills. You can register public, private, or local catalogs of your own with repertoire catalog add. See Catalogs and Private and company catalogs.

Catalog trust¶

The optional trust block on a catalog registration. It lists armored public keys, next to the manifest, and the fingerprints those files must match. When the block is present, Repertoire accepts the catalog only if the checked-out commit has a valid signature from one of those keys, checked in a temporary keyring, and records that key's fingerprint in the lock. A registration without the block skips the check and omits the fingerprint. See Trusted catalog keys.

Skill digest signature¶

REPERTOIRE.digest.asc, a detached signature over a skill's content digest. Trusted catalogs require this file in the skill directory. The digest is the hex hash of the skill's paths, modes, symlink targets, and file bytes, and the signature file is not part of that hash. Either key in the catalog trust block may sign it. A trusted install records that key's fingerprint in the lock. Loose catalog skills use the same file. See Skill digest signatures.

Loose catalog¶

A Git repository (or local folder) of SKILL.md directories with no repertoire.yaml catalog section. Repertoire discovers skills under a bounded walk (skills trees, agent skill dirs, and marketplace layouts such as plugins/<plugin>/skills/<skill>) and synthesizes a catalog in memory so you can still register, install, and lock it. A written catalog manifest is what unlocks variants, project instructions, hooks, and stubs.

Target¶

One agent's skills folder, named after the agent: codex, claude, cursor, gemini, and so on.

--target all means every supported agent. With no --target, add picks the agents it detects on your machine. The full list of names and install paths is in Targets and security.

Scope¶

Where a skill lives. Global (the default) installs under your home directory, so the skill is available in every project. Project (--project) installs inside the current Git repository, for skills that should ship with that repository.

Manifest¶

The repertoire.yaml file that says what you want installed. In a project it lists the catalogs and skills contributors need; in your user configuration directory it records the skills you added globally. A catalog also uses a repertoire.yaml, with a catalog: section, to publish skills. See Manifests and state.

Lock file¶

repertoire.lock.json, written by Repertoire next to the manifest. It records exactly what was installed: catalog commit, content digest, and target paths. When the catalog has a trust block, it also records the fingerprints that signed that commit and the skill digest. The lock file is how Repertoire notices when you have edited an installed copy. Never edit it by hand.

Managed¶

An installed copy that Repertoire owns and tracks in the lock file. Updates and removals refuse to touch a managed copy you have modified locally unless you pass --force.

Source-qualified ID¶

A skill name prefixed with its catalog's host and path, such as github.com/phillarmonic/ai-skills/zensical. It names the catalog and the skill together, so it is never ambiguous when two catalogs define the same short name.

Provenance¶

Where an installed skill came from and whether each managed copy still matches the lock: catalog name, source, commit, content digest, targets, and intact / modified / missing status. repertoire show <skill> prints it.

Dry run¶

--dry-run on a mutating command prints the writes and refusals that would happen, and performs no disk, lock, or manifest changes.