Codex runtime mapping

This published workflow is explicit-only through agents/openai.yaml and bounded to local design work with observable source grounding, at least two structurally distinct candidates, an independent judge, and disposable verification. If the grounding or competition gate cannot be met, report PARTIAL or unavailable rather than a completed Architect result. This workflow depends on cursor-how, cursor-why when rationale matters, and cursor-arena for structural alternatives. The generated cursor-how/cursor-why mirrors and their relevant references have been read in a bounded native follow-up; the receipts preserve paths, outputs, nulls and gaps. Resolve exact installed names before invocation; a published mirror and an observed read/role-flow are not proof of automatic availability or triggering. In the observed native path, candidate delegates used collaboration.spawn_agent with agent_type=general-worker, the cross-judge used agent_type=reviewer, and the failed-hypothesis redesign used agent_type=planner; these native roles are capability mappings, not proof of the upstream Cursor runner slugs or automatic trigger. Native model and effort selection follows the standing model-capability-router; invocation metadata and worker self-report must be recorded separately. The coordinator owns design synthesis, implementation and re-design when the sketch fails.

Architect

Sketch types, function signatures, class shapes, and module boundaries with not implemented bodies and pseudocode. Synthesize across multiple model perspectives, then fill in code against the chosen sketch.

Start

Open a todolist with one entry per phase before starting.

1. Ground
2. Sketch
3. Agree
4. Implement
5. Scrap

Phase A: Ground the problem