autoresearch-pro

Overview

Automatically improve any OpenClaw skill, prompt, or article through iterative mutation-testing: small edits → run test cases → score with checklist → keep improvements, discard regressions.

Inspired by Karpathy/autoresearch.

Supports three optimization modes:

Workflow

Step 1 — Identify Mode and Input

Ask the user to confirm:

• Mode 1 — Skill: User says "optimize [skill-name]" or provides a skill path
• Mode 2 — Prompt: User says "optimize this prompt" or pastes a prompt
• Mode 3 — Article: User says "improve this article" or pastes article text

For Skill mode, resolve the skill path to ~/.openclaw/skills/\x3Cskill-name>/SKILL.md. For Prompt/Article mode, keep the text in context (do not write to disk unless needed).

Step 2 — Generate Checklist (10 Questions)

Read the target content first. Then generate 10 diverse, specific yes/no checklist questions relevant to the content type:

For Skill mode (same as before):

For Prompt mode (10 tailored questions):

For Article mode (10 tailored questions):

Present the 10 questions, numbered 1-10. Ask the user to select which ones to activate (e.g., "use questions 1, 3, 5, 7"). Default: use all 10 if user doesn't specify.

Step 3 — Prepare Test Cases

• Skill mode: Generate 3-5 realistic prompts a user would send when using the skill
• Prompt mode: Generate 3-5 test inputs that the prompt would process
• Article mode: Generate 3-5 ways the article might be read or consumed

Store test cases in context — do not write to disk.

Step 4 — Run Autoresearch Loop

Loop configuration:

• Rounds per batch: 30
• Max total rounds: 100
• Pause: After every 30 rounds, show summary and ask user to continue or stop
• Stop conditions: User says stop, OR 100 rounds completed

Per-round procedure:

1. Mutate: Make ONE small edit to the target content:

   • Skill mode: edit SKILL.md
   • Prompt mode: edit the prompt string
   • Article mode: edit the article text

2. Test: For each test case, simulate what output the content would produce.

3. Score: Apply each active checklist question (0 or 1 per question). Score = (passed / total) × 100.

4. Decide: If new score ≥ best score → keep the mutation. If lower → revert.

5. Log: Round number, mutation type, score, keep/revert decision.

Mutation types (pick one per round):

Step 5 — Report Results

After each batch (30 rounds):

Batch N (rounds X-Y):
  Best score: XX%
  Mutations kept: N  |  Reverted: N
  Most effective types: [list top 2-3]
Accumulated improvements: [summary]
Continue? (yes/stop)

After full completion:

• Original score vs. final score
• Top 3 most impactful mutations
• Final improved content (inline or diff)
• File path (skill mode only)

Mutation Strategy Reference

High-impact, low-risk changes:

• Adding explicit constraints where the content is vague
• Expanding coverage to cover edge cases
• Adding concrete examples to abstract instructions
• Tightening soft language ("try to" → "must")

Avoid in one round:

• Large rewrites of entire sections
• Multiple unrelated changes at once
• Changing fundamental scope or purpose

See references/mutation_strategies.md for the full strategy guide.

Mode Selection Quick Reference

Default to Prompt mode if the input is a text string without a skill path.