Skip to content

10-agent-integration-protocol.md

Latest commit

 

History

History
28 lines (18 loc) · 1.17 KB

10-agent-integration-protocol.md

File metadata and controls

28 lines (18 loc) · 1.17 KB

Agent Integration Protocol

This workstream is about integrating additional agents into the existing automation ecosystem.

The current ecosystem is anchored in stranske/Workflows and its consumer repos.

Required deliverables

  • Claude Code integrated as the next agent
  • A third agent integrated (Aider / Continue / approved alt)
  • A stable “outputs contract” for agent runs so keepalive/verifier does not require per-agent special cases
  • Documentation that enables a future agent to be added by following the docs (docs-first test)

Ground truth file references (Workflows)

  • Agent control plane catalog and topology: ../.workflows-lib/docs/ci/WORKFLOWS.md in Workflows
  • Orchestrator entry point: .github/workflows/agents-70-orchestrator.yml
  • PR keepalive detection: consumer .github/workflows/agents-pr-meta.yml calling Workflows reusable-20-pr-meta.yml

Contract expectations

Any agent integration should:

  1. Emit a machine-readable result object (pass/fail + summary + artifacts)
  2. Produce stable filenames for prompt/output artifacts
  3. Respect the “no secrets in logs” rule

If it’s not testable in Workflows-Integration-Tests, it’s not integrated.