AI & Agents
Agent Catalog
An agent is a reusable definition of AI-agent work: a prompt, an engine, an input schema, and the set of tools the definition asks for. Agents live in your organization's catalog, and any number of bindings — an Agentic Job on a schedule, a manually triggered run — can reuse the same definition without re-authoring or re-approving it.
Navigate to Organization Settings → Agents to browse the catalog. Each entry shows its engine, visibility, status, and where its definition lives.
Plan Availability
Agents are available on plans that include AI features, and individual engines may be gated separately. Contact your CodeCargo administrator if the catalog or a specific engine is not visible in your organization.
Visibility
| Visibility | Description |
|---|---|
| Private | Created implicitly by a single job and usable only by that job |
| Org | Published to the catalog, reviewed by an admin, and available for anyone with access to bind to a job |
When you create an Agentic Job and write its prompt inline, CodeCargo records a private agent for you automatically — nothing extra to configure. Use Save as reusable agent on the job to promote that definition into the catalog so other jobs can bind it.
Publishing and Tool Grants
A definition requests capability; an admin grants it. Publishing an agent opens a review form where an Org Admin picks which of the requested tools the agent is actually allowed to use — the grant can only be equal to or narrower than what the definition requested, and it is pinned to the exact revision that was reviewed.
The tools an agent can actually call on a given run are always the intersection of what the agent was granted and what the binding allows. Narrowing either side narrows the run; neither side can widen the other.
Pinned revisions for state-changing tools
An agent that holds tools which change state — pushing commits, opening pull requests — must be pinned to a specific commit rather than a branch or tag, so the reviewed definition cannot change underneath the grant.
A tool can also be granted with approval required. The agent may call it, but the run pauses until a human approves that specific call — see Agentic Jobs → Tool Approvals.
Committing an Agent to a Repository
Agents can be stored in git as .codecargo/agents/<name>.agent.md, so the definition is reviewed and versioned like any other code. From an agent's detail page, click Commit to repo to open a pull request that adds the file to the repository and branch you choose.
When that pull request merges, the organization sync picks the file up and attaches it to the existing catalog entry rather than creating a duplicate — the same agent simply becomes git-backed. From then on, edits to the file flow into the catalog on the next sync, and changes still need a fresh grant before they take effect.
Engines
Every catalog entry is labelled with the engine that runs it, and engines are governed the same way regardless of who executes them:
- CodeCargo — the built-in code-editor agent that runs in a CodeCargo workspace
- GitHub Agentic Workflows —
.github/workflowsagentic definitions detected during organization sync and imported into the catalog (see Agentic Workflows) - GitHub Copilot — Copilot custom and system agents synced from your GitHub organization
GitHub-executed engines are callable from CodeCargo where your plan includes them; their runs appear alongside CodeCargo runs.