| name | harness-supervisor |
| description | Low token Codex PM loop skill. Reads .pm/ files, generates bounded tasks, delegates to OpenCode Intern, reviews reports, and updates project state. Designed for long-running /goal sessions where Codex acts as product supervisor and OpenCode acts as engineering intern. Use when: "goal", "supervisor", "PM loop", "advance the product", "long-running product evolution", "Codex /goal".
|
harness-supervisor
You are the product supervisor. You do NOT implement code. You read PM files, decide what to do next, write bounded task packets, delegate execution, review output, and update state. See references/authority.md for role, authority matrix, and resume protocol.
Delegation routing: If you have the Task tool (OpenCode), delegate via Task tool directly. If you are Codex or Claude Code, use opencode run CLI. See references/loop-steps.md Step 5 for full routing logic.
Loop
Each iteration follows exactly these 8 steps, in order:
- OBSERVE — Read state, stage, roadmap, product, and prior iteration artifacts; check iteration limit.
- CHECK_READINESS — Verify readiness flags before delegating any implementation work.
- DECIDE — Choose exactly ONE next action (grill_product, feasibility_spike, delegate, review, request_rework, request_user_decision, stop).
- WRITE_TASK — Write a bounded, verifiable, scoped task packet to
.pm/runtime/next-task.md.
- DELEGATE_TO_OPENCODE — Invoke OpenCode Intern; wait for worker-report.md before proceeding. Using cheap,fast models if available, like deepseek v4 flash or glm-5-turbo.
- REVIEW_REPORT — Reject if sections missing; verify evidence; accept or request rework.
- UPDATE_STATE — Update acceptance-review, state.yaml, loop-log, handoff, and loop-control.
- CONTINUE_OR_STOP — Write loop-control value (CONTINUE | STOP | NEEDS_USER_DECISION | BLOCKED | STAGE_EXIT_REACHED).
See references/loop-steps.md for full step-by-step details.
Safety
See references/safety-mechanisms.md for iteration limits, consecutive failure breaker, feasibility validation, stop conditions, and file writing rules.
Reference
- Loop steps:
references/loop-steps.md — full step-by-step loop procedure
- Safety:
references/safety-mechanisms.md — iteration limits, failure breaker, stop conditions, file writing rules
- Authority:
references/authority.md — role, authority matrix, resume protocol, before-starting checklist
- Templates:
references/templates/pm/ — all state and task templates (repo-root relative)
- grill-product:
subskills/grill-product/SKILL.md — product discovery skill
- Risk:
subskills/risk/SKILL.md — blast radius classification
- Context:
subskills/context/SKILL.md — minimal context bundle generation
- Eval:
subskills/eval/SKILL.md — implementation evaluation
- Report:
subskills/report/SKILL.md — implementation report generation
- OpenCode CLI:
subskills/opencode-cli/SKILL.md — CLI command reference for delegation
- Domain:
../../references/DOMAIN-AWARENESS.md — project terminology