Skip to content

iteration-prompt.md

Latest commit

 

History

History
91 lines (62 loc) · 2.05 KB

iteration-prompt.md

File metadata and controls

91 lines (62 loc) · 2.05 KB

Autoagent Iteration Prompt

You are running an optimization iteration for the autoagent skill.

Context

Read these files from the sandbox:

  • current-guidance.md - The guidance being optimized
  • scores.md - History of scores and changes
  • scoring.md - How to measure success
  • fixtures/test-cases.json - Test inputs

Your Task

Step 1: Analyze Current State

  • Read current guidance
  • Review score history (last 10 runs)
  • Identify patterns in what's been tried
  • Note current score

Step 2: Propose Edit

Generate ONE specific edit to the guidance that might improve the score. The edit should:

  • Be specific and actionable
  • Address a weakness identified in scoring
  • Not be identical to recently tried changes

Format:

## Proposed Edit

**Rationale:** Why this change might help

**Change:**

[Show exact diff or new text]

Step 3: Apply Edit

Write the edited guidance to current-guidance.md

Step 4: Run Test

Use a subagent to run the task with the new guidance:

  • Give the subagent current-guidance.md
  • Provide test inputs from fixtures/test-cases.json
  • Capture the output

Step 5: Score Result

Evaluate the output against scoring.md criteria. Generate a score 0-100.

Step 6: Log Decision

Append to scores.md:

| N   | Description of change | SCORE | keep/discard |

Where N is the run number (increment from last).

Step 7: Update Guidance

  • If score improved: Keep the edit (current-guidance.md is already updated)
  • If score declined: Revert current-guidance.md to previous version

Step 8: Check Plateau

If last 10 scores are identical:

  • Log "Plateau detected - pausing"
  • Notify user
  • Stop the cron (or pause and await user override)

Output Requirements

After completing:

  1. Updated scores.md
  2. Possibly updated current-guidance.md (if change kept)
  3. Brief summary of what happened

Remember

  • Be specific in edits - vague changes won't score well
  • Learn from score history - don't repeat failed approaches
  • Non-destructive: original guidance stays in guidance-under-test.md