skills.md
Are you an agent?

skills.md documentation

Build agent workflows on remote skills.

Install the CLI, connect your agent, discover public and premium skills, and return files from remote runs without copying private implementation source.

MCP Reference

The MCP server exposes the same registry, docs, quote, run, and artifact surfaces as the CLI. Agents should use MCP for discovery and execution instead of reading local skill folders.

GETskills://registry

Registry resource

Returns public registry metadata for skills the agent can discover. This is enough to choose a skill without loading every skill document into context.

POSTlist_skills

List skills

Lists skills by profile, category, or tag. Use this before asking for a specific skill document.

POSTget_skill_docs

Fetch selected docs

Returns the instruction contract for one selected skill. The response should be concise and source-free for hosted premium skills.

POSTquote_skill

Quote a paid run

Estimates the user-visible cost for premium work. Provider routing and internal margins stay server-side.

POSTrun_skill

Submit a run

Starts a skill run after auth and approval checks. Long jobs return a run id that agents can poll.

POSTdownload_artifacts

Materialize outputs

Downloads run artifacts into .skills/exports/<skill>/<run-id>/ without writing private implementation source.

Agent behavior

Discover remotely

List skills or search the registry. Do not scan .claude/skills, .codex/skills, or .gemini/skills for private skill source.

Load one contract

Fetch docs for the one skill needed for the task. Keep the rest of the catalog out of context.

Quote and ask approval

For paid or sensitive work, show the quote and wait for approval before running.

Return file paths

After completion, give the user the generated artifact paths and receipt metadata.

bun install -g @hasna/skills