メニュー
Manages Ralph Wiggum Loop state files (TASKS.md, PROGRESS.md, LOOP_CONFIG.md). Use this skill when asked to initialize loop state, create task files, reset progress, compact progress logs, or manage the loop's persistent disk state.
Codex または Claude でインストール この Prompt をコピーして Codex、Claude、または他のアシスタントに貼り付けると、Skill ページを確認してインストールできます。
SOC 職業分類に基づく
Enforces safety guardrails for the Ralph Wiggum Loop: iteration limits, path restrictions, diff size limits, and validation gates. Use this skill when checking if a loop operation is within safety bounds or when configuring loop safety parameters.
Runs the Ralph Wiggum Loop — an iterative fresh-context agent loop. Use this skill when asked to execute a loop, start iterating on tasks, or run the ralph wiggum pattern. Provides the loop execution script and instructions for invoking it.
Detects convergence issues in Ralph Wiggum Loop iterations: oscillation, stagnation, metric gaming, and context overflow. Use this skill when checking loop health or when the loop seems stuck.
| name | task-state-manager |
| description | Manages Ralph Wiggum Loop state files (TASKS.md, PROGRESS.md, LOOP_CONFIG.md). Use this skill when asked to initialize loop state, create task files, reset progress, compact progress logs, or manage the loop's persistent disk state. |
This skill manages the persistent state files used by the Ralph Wiggum Loop.
The loop uses three core state files, all in the project root:
| File | Purpose | Created by |
|---|---|---|
TASKS.md | Task list with status markers | User or loop-planner agent |
PROGRESS.md | Append-only iteration log | Loop orchestrator |
LOOP_CONFIG.md | Loop configuration | User or init-state.sh |
Run init-state.sh to create template state files:
bash .github/skills/task-state-manager/init-state.sh [--force]
Use --force to overwrite existing files.
Tasks in TASKS.md use these status markers:
| Marker | Meaning |
|---|---|
[ ] | Pending — not yet started |
[~] | In progress — being worked on |
[x] | Done — completed successfully |
[!] | Failed — attempted but unsuccessful |
[STOP] | Stop flag — halt the loop |
When marking a task complete:
### 1. Implement user model
- **Status:** [x]
- **Completed:** 2024-01-15T10:30:00Z
When marking a task failed:
### 3. Fix auth middleware
- **Status:** [!]
- **Error:** TypeError: Cannot read property 'token' of undefined
- **Attempted:** 2024-01-15T11:00:00Z
If PROGRESS.md grows too large (>500 lines), compact it:
To reset for a new loop run:
TASKS.md but reset all statuses to [ ]PROGRESS.md to PROGRESS.{timestamp}.mdLOOP_CONFIG.md as-isPROGRESS.md — it's the audit trailPROGRESS.md, never overwrite existing entriesTASKS.md is the source of truth for what work remainsLOOP_CONFIG.md controls safety limits — don't weaken them during a loop