# Self Refine Loop

> Runs a generator-critique-reviser loop to iteratively improve agent outputs. Use for: self refine, reflexion loop, critique and revise, improve output.

- Skill: `stijnman/self-refine-loop` (Agent Skill)
- Install (CLI): `npx skillmds@latest add stijnman/self-refine-loop`
- Raw SKILL.md: https://api.skillmd.com/api/skills/stijnman/self-refine-loop/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: AI & ML
- License: MIT
- Author: Stijnman (https://skillmd.com/u/stijnman)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/stijnman/self-refine-loop

---

# Self Refine Loop
## When to Use

- User says **self refine** or task matches this capability
- User says **reflexion loop** or task matches this capability
- User says **critique and revise** or task matches this capability
- User says **improve output** or task matches this capability

## Workflow

1. Capture the current output and the user's quality criteria.
2. Generate a critique listing specific weaknesses (max 5 bullets).
3. Revise the output addressing every critique point.
4. Score confidence 0-10 on whether criteria are met.
5. Repeat until confidence >= 8 or 5 iterations; return best version with changelog.

## Integrations

- `goal-verifier`
- `agentic-uncertainty-quantifier`
- `dspy-prompt-optimizer`

## Error Handling

| Failure | Response |
|---------|----------|
| No criteria given | Ask user for 1-3 success criteria before looping. |
| Confidence stuck below 5 | Stop early; report blocker and ask for guidance. |
| Output grows unbounded | Cap revisions to prior length + 20%. |

## Gotchas

- Do not loop on trivial typos; one-pass fix is enough.

## Example

**Input:** User request matching triggers above.
**Output:** Structured result per workflow with integrations invoked as needed.