Skip to main content
Last updated: 2026-09-06
orca skills manages the tenant skill catalog: Agent Skills packages (a SKILL.md plus any scripts, references, and assets) that can be attached to agent profiles. A skill’s source is user, imported, or platform. Every subcommand accepts the global flags. All positionals on this page are required; there is no interactive picker. Destructive subcommands prompt for confirmation in a terminal; in a script --yes is required or the command refuses (exit code 2).

​
orca skills list

Lists the skills in the catalog.
Plain (piped) output is one tab-separated row per skill: name, source, resource file count, description.

​
orca skills get

Shows one skill. With --json the full skill object is printed (name, description, body, tags, source, resources, license, compatibility, requiresSandbox, and so on). The plain view prints key and value rows: name, source, description, tags, resource count. The terminal view adds the SKILL.md body. With --resource, stdout carries the file contents exactly as stored, so it can be redirected to disk. With --resource and --json together:

​
orca skills delete

Deletes a skill from the catalog.

​
orca skills import

Imports an Agent Skills folder into the catalog. <path> must be a directory with a SKILL.md at its root; a single file, an archive, or a directory without SKILL.md exits 2 before anything is uploaded. Every file under the directory is uploaded recursively with paths relative to that directory, so the directory’s own name does not have to match the skill’s name. Import is two-phase. The package is first uploaded for validation: validation errors are printed to stderr as error: ... and the command exits 2; warnings are printed as warning: ... and do not block. With --dry-run the command stops there and reports the skill name, resource file count, total bytes, and whether the skill requires a sandbox (it bundles scripts). Otherwise the staged package is committed to the catalog. Importing a name that already exists is a usage error (exit code 2) whose message points at --force; on cli-v0.5.0 the same conflict exits 1. With --json and --dry-run the preview object is printed: skill (the parsed SKILL.md frontmatter and body), resources, validation (ok, errors, warnings), requiresSandbox, totalBytes, and the stagingId the commit would use. With --json alone the created skill record is printed.
A skill that requires a sandbox only works on agents that have one enabled; see Skills.

​
orca skills attach

Attaches a catalog skill to an agent profile. The profile is read first, so an unknown agent exits 4. If the skill is already attached nothing is written and changed is false. A skill that is not in the catalog exits 2 with a hint to import it; a server that does not support modifying skills on the agent exits 1.

​
orca skills detach

Detaches a skill from an agent profile. The profile is read first, so an unknown agent exits 4. If the skill is not attached nothing is written and changed is false.
Verified against orca-cli cli-v0.5.0.