planning-supervisor

name	planning_supervisor
description	Orchestrate a lightweight plan review loop (Planning → Critic → Planning → Verifier) with iteration budget and optional hooks.

Planning Supervisor Skill (MVP)

Purpose

Coordinate a fixed planning loop that produces a single final plan.md artifact and (optionally) supporting review artifacts.

Inputs

• Task intake (goal/context/constraints/acceptance criteria)
• debug_mode: on|off (default: off)
• iteration_budget: integer (default: 2)

Outputs

• plan.md (final PlanArtifact)
• critique.md (only if debug_mode=on)
• verify.md (only if debug_mode=on)

Orchestration Contract

You MUST run the following loop in order:

1. Intake normalization
2. Generate PlanArtifact v0 (delegate to planning)
3. Generate CritiqueArtifact (delegate to plan_critic)
4. Revise to PlanArtifact v1 (delegate to planning with critique)
5. Run Plan-lint and produce VerificationReport (delegate to plan_verifier)

Stop conditions:

• Stop if:
  • no Blocker and no High in critique
  • verifier returns plan_lint = passed
  • iterations ≤ iteration_budget
• If critique has Blocker or High OR verifier returns failed:
  • attempt one more revision (within budget)
  • if still failing or budget exceeded: output 1–5 clarifying questions (only those that block correctness)

Hook Points (MVP: define, do not execute)

You MAY mark escalation requests in the plan’s Risks section, but MUST NOT run extra critics in MVP.

Hooks (future):

• SecurityCritic (auth/permissions/secrets/external IO/file upload)
• DataMigrationsCritic (schema/migrations/mass updates/backups)
• TestingCritic (critical layer changes without tests)
• DeployOpsCritic (env/CI/CD/release flags)