Agents, Skills & Knowledge
The shared catalog of reusable agent roles, skills, and knowledge resources that workflows reference.
Alongside workflows, the Hub maintains a shared catalog of reusable building blocks. They live in the database, are readable by operator tokens (api, mcp, read, approvals, or admin), and are edited only by admins — so every workflow on the Hub draws from the same curated set.
The three catalogs
| Catalog | What it is |
|---|---|
| Agents | Reusable agent roles: a slug, name, description, instructions, a tool list, linked skill slugs, and tags. A role captures "how this kind of agent should behave" once, instead of restating it per workflow. |
| Skills | Named, versioned instruction bodies (a slug, description, and body text) that agents and workflows can pull in by slug. |
| Knowledge | Reference resources — documents or links with a type, body/URL, and tags — searchable with ?q= / search_knowledge. |
All three are versioned and can be individually enabled/disabled.
How workflows reference them
A workflow definition declares requiredAgents and requiredSkills as lists of catalog slugs; describe_workflow reports them alongside the schemas so a caller can see what a workflow depends on before running it. Workflow package exports record these requirements, and import preview reports whether the target Hub satisfies them.
Who can edit
Reading any catalog requires an operator-readable token. Runner-only tokens are limited to the runner protocol and cannot inspect the catalog. Creating and updating entries requires the admin scope — the catalogs are part of the Hub's trusted configuration, like secrets and hook profiles.
# Browse the catalogs
curl -H "Authorization: Bearer $RUNYARD_TOKEN" https://hub.example.com/api/agents
curl -H "Authorization: Bearer $RUNYARD_TOKEN" https://hub.example.com/api/skills
curl -H "Authorization: Bearer $RUNYARD_TOKEN" "https://hub.example.com/api/knowledge?q=release"In the web app the three catalogs share one section with deep links #agents/agents, #agents/skills, and #agents/knowledge (see Deep links).
API & MCP
| Endpoint | MCP tool | Notes |
|---|---|---|
GET /api/agents | list_agents | |
POST /api/agents | create_agent | Admin. |
PATCH /api/agents/{slug} | update_agent | Admin. |
GET /api/skills | list_skills | |
POST /api/skills | create_skill | Admin. |
PATCH /api/skills/{slug} | update_skill | Admin. |
GET /api/knowledge | search_knowledge | ?q= searches. |
POST /api/knowledge | create_knowledge | Admin. |
PATCH /api/knowledge/{slug} | update_knowledge | Admin. |