@in-org-quicko/skillset-cli

Command-line client for publishing and managing Skills on a Skillset Registry.

Install

npm install -g @in-org-quicko/skillset-cli

This installs a skillset binary.

Usage

skillset login --registry <url>
skillset search [query]      # what the team has published
skillset info <name>         # read a Skill without installing it
skillset install <name>      # install a Skill for a coding agent
skillset install <url> --name <skill>  # install straight from a GitHub or GitLab repo
skillset list                # what's installed here, and whether it's current
skillset update [names...]   # re-download whatever the Registry has moved on from
skillset remove <name>       # uninstall a Skill
skillset publish [path]      # defaults to the current directory
skillset publish <url> --name <skill>  # publish straight from a GitHub or GitLab repo
skillset whoami

Without a global install, every command runs through npx:

npx @in-org-quicko/skillset-cli install <name>

Namespaces

Two parties can publish the same Skill name — the team's own pdf and one imported from anthropics/skills. A bare name still reads: it resolves to the only match, or to the one published to your Registry. Only a tie between two outside parties has to be spelled out.

skillset search pdf                             # imported rows show their namespace
skillset install pdf                            # yours, or the only one
skillset install pdf --namespace anthropics/skills

A project holds one .agents/skills/<name>, so installing a Skill over one a different party named is refused; pass --force to replace it.

Installing from a repository

skillset install https://github.com/acme/skills --name pdf
skillset install https://github.com/acme/skills/tree/main/pdf --name pdf   # or a folder in it

The CLI clones the repository with your own git, so any repository you can clone works — private ones included, over HTTPS or SSH, with whatever credentials git already has. It needs git on PATH. --name is required: it picks the Skill by the name in its SKILL.md.

The Skill is recorded in skillset-lock.json with the repository it came from. If you are logged in, it is also submitted to the Registry for an Admin to approve (Settings → Submissions). Until then skillset update reports it as pending; once approved, update switches it to the Registry's copy.

Publishing from a repository

skillset publish https://github.com/acme/skills --name pdf

Exactly like install <url>: the repository is cloned with your own git, so private ones work too, and --name (required) picks the Skill by the name in its SKILL.md. The Skill records the repository as its Source, and takes its Namespace from it. A Skill published from a path records no Source and reads as published straight to your Registry.

Credentials are stored per-Registry after login. For CI, skip login and set:

SKILLSET_REGISTRY=<url>
SKILLSET_TOKEN=<token>

SKILLSET_REGISTRY and SKILLSET_TOKEN always override the stored config.

Machine-readable output

Every command takes --json, which replaces the formatted output with the command's result on stdout:

skillset search code-review --json | jq -r '.items[].name'
skillset list --json | jq '[.skills[] | select(.status == "outdated")]'

Failures go to stderr as { "error": { "code, message, status?, field? } } with a non-zero exit code, so a pipeline only ever parses well-formed payloads. code is the Registry's own error code where there is one, and cli_error for a failure that never reached it.

--json never prompts, whatever the terminal says — a clack prompt would write to the same stdout the payload goes to. So login needs --token, a multi-Skill publish needs --yes, and install needs --agent/--scope where it would otherwise have asked.

License

AGPL-3.0-only