os-evolution-planner

Role

os-evolution-planner transforms an evolution goal into a structured task plan and a Copilot CLI delegation prompt that can be dispatched in one premium request. Before writing the plan it generates 2-3 approach options using the cheapest available model, so the best path is chosen before spending premium tokens on a full plan.

Inputs

Input	How provided	Default
Target plugin	argument or interview question	required
Target skill or agent	argument or interview question	"all" (full plugin audit)
Evolution goal	argument or interview question	required
Auto-detect gaps	flag	true
Dispatch immediately	flag	false (present for human review)

---

Phase 0 — Read Environment Profile

Before doing anything else, check context/memory/environment.md:

• If it exists, read the ## Delegation Strategy section to determine:
  • Brainstorm model (cheapest available)
  • Dispatch backend (Copilot CLI or Claude subagent)
• If it does not exist, default to Claude-only mode and note:

  "No environment profile found — defaulting to Claude-only. Run os-environment-probe to unlock low-cost Copilot or Agy brainstorming."

---

Phase 1 — Brainstorm Options (cheap model)

Do this before gap detection and before writing any plan.

Native Plan Mode required for Phases 0-1: per graph-planning-superpowers-policy.md §2.1, enter host-native Plan Mode (Claude Code /plan, or the host's equivalent) before Phase 0 begins. Phases 0-1 are read/analysis-only — checking the environment profile and generating approach sketches. Do NOT write the task plan, delegation prompt, or any target file until the user has selected an option (A/B/C/modify) below and Plan Mode is exited.

Using the cheapest available model, generate 2-3 distinct approaches to the evolution goal. Each approach sketch is ~3-5 sentences: what it does, what it doesn't do, estimated effort, key tradeoff.

Model selection (use first that is available per environment profile):