Skip to content

How to onboard an agent repository

Quickstart

See Quickstart

Reading Order

  1. Onboard an agent to EAP — overview, two-plane model, repo structure, before-you-start.
  2. Write AGENTS.md — define the agent, its routing, and subagents.
  3. Author skills — write the SKILL.md files referenced from the agent.
  4. Plan MCP dependencies — inventory the external systems your skills need.
  5. Register with enterprise-agent-platform-registry — wire the agent repo, MCPs, and access lists into a project YAML.
  6. Test and promote — local Codex → EAP staging → EAP production.
  7. Change lifecycle — how the two control planes interact during and after rollout.
  8. MCP catalog — what your project can request in mcp:.

What You Need To Answer Before You Start

Before writing anything, you should be able to answer:

  • what outcome does the agent produce, and for which users
  • which Cisco MyID group(s) and CI client IDs should be allowed to call it
  • which MCP servers does it need (read-only vs state-changing)
  • what evidence must appear in every result
  • what conditions should pause for approval, and which should refuse outright
  • who owns the dependent systems if a tool call fails

If those answers are vague, the project isn't ready for EAP review yet.