# Gsp Rolling Supervisor

> Use when game work is too large for one session and needs a guarded external build/verify/repair loop.

- Skill: `mike007jd/gsp-rolling-supervisor-2` (Agent Skill, multi-file: 10 files)
- Install (CLI): `npx skillmds@latest add mike007jd/gsp-rolling-supervisor-2`
- Raw SKILL.md: https://api.skillmd.com/api/skills/mike007jd/gsp-rolling-supervisor-2/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Coding & Dev Tools
- Author: mike007jd (https://skillmd.com/u/mike007jd)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/mike007jd/gsp-rolling-supervisor-2

---


# Game Rolling Supervisor

## Goal
Push a game project through repeated autonomous passes without letting one long context rot into uselessness.

## Outputs

Follow the `gsp-orchestrator` output strategy:
- **inline** (default): present rolling supervisor plan in conversation.
- **minimal** or **full**: write `docs/game-studio/rolling-supervisor.md` and `docs/game-studio/rolling/state.json`.

Use:
- `./shared/templates/rolling-supervisor-plan.md`
- `./shared/reference/rolling-supervisor-architecture.md`
- `./schemas/rolling-state.schema.json`
- `./schemas/session-result.schema.json`

## Rules
- use this only when the repo can tolerate repeated build/verify/repair cycles
- prefer **short-lived worker sessions** over one giant session
- every worker must emit a compact structured handoff
- verification gates decide whether to continue
- use worktree isolation for parallel branches when practical
- do not create infinite hook or stop loops
- if the project is live or risky, only enable with explicit authorization

## Worker graph
Typical graph:
1. concept/requirements brainstorm
2. scope profile
3. architecture/plan
4. build
5. verify
6. feedback design/polish
7. verify again
8. repair if needed

## Important
The supervisor belongs outside the plugin runtime.
The plugin defines the contracts; an external Python supervisor executes them.