# Refine

> Sharpen todo items in brain/todos.md into precise, actionable prompts via targeted questioning. Triggers: "refine", "refine todos", "clarify todos".

- Skill: `poteto/refine` (Agent Skill)
- Install (CLI): `npx skillmds add poteto/refine`
- Raw SKILL.md: https://api.skillmd.com/api/skills/poteto/refine/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Productivity
- Author: poteto (https://skillmd.com/u/poteto)
- Updated: 2026-09-09
- Page: https://skillmd.com/skills/poteto/refine

---


# Refine

Read `brain/todos.md` and refine unclear items into actionable prompts through targeted questioning.

## Workflow

**Use Tasks to track progress.** Create a task for each step below (TaskCreate), mark each in_progress when starting and completed when done (TaskUpdate). Check TaskList after each step.

1. **Read** `brain/todos.md` and review items under `## Open`.
2. **Triage** each item: is it clear enough to act on without further context? A clear item specifies *what* is wrong, *where* it happens, and *what the fix should be*. Skip items that already meet this bar.
3. **Ask** clarifying questions about unclear items using `AskUserQuestion`. Group related questions — aim for one round of questions, not a back-and-forth (to minimize back-and-forth and user fatigue). Good clarifying questions target:
   - **What** — what exactly is the problem or desired behavior?
   - **Where** — which part of the app, which component, which flow?
   - **How** — how should it work instead? What does "fixed" look like?
4. **Rewrite** unclear items in-place using `Edit`, incorporating the user's answers. Each rewritten item should be a self-contained prompt that another agent could act on without additional context.
5. **Remove** any items the user says are already done or no longer relevant.
6. **Report** a summary of what changed.