mcs-pipeline-test

name

mcs-pipeline-test

description

Use this skill to run the MCS build pipeline verify-fix harness against the canonical kitchen-sink agent. The harness reads back every component from MCS via API, classifies any drift against a fixed taxonomy, and auto-applies fixes for confirmed classifications. Use when validating the build pipeline end-to-end, after touching tools/mcs-lsp.js, tools/island-client.js, tools/add-tool.js, or app/lib/build-runner.js, or when component pushes are silent-failing in real builds.

MCS Pipeline Verify-Fix Harness

Builds the canonical "kitchen-sink" test agent in the dktest environment, reads back every component from MCS, and iterates fix → verify until all verifiable components match the spec. Wraps /mcs-build; does not replace it.

This skill is the entry point for the harness. The runner that drives the loop is tools/pipeline-test-loop.js. The verifier library is tools/verify-component.js. Failure taxonomy lives in tools/pipeline-classifications.js. Fix planner lives in tools/pipeline-fix-map.js.

Smoke-phase coverage (current): only verifyTopic is wired live. Other component verifiers return status: 'skipped' and surface as "not yet implemented." Each iteration of harness work expands one verifier at a time. When a verifier ships, its corresponding classification becomes auto-fix-eligible if pipeline-fix-map.js lists it as type: 'auto'.

Out of scope for this skill: building real customer agents (use /mcs-build), evaluating an agent's behavior (use /mcs-eval), and Playwright UI snapshots of the MCS portal (deliberately not implemented — see plan).

---

Pre-flight

Before invoking the runner, confirm the local environment is set up correctly. These checks are fast and prevent wasted iteration on auth/wrong-env issues: