schliff

# Schliff — Skill Measurement & Iteration Framework Constraint + clear metric + disciplined iteration = compounding gains. The composite score measures structural quality (file organization, keyword coverage, eval suite breadth) — not runtime effectiveness. Use `--runtime` to validate actual behavior. ## Quick Start (Only 2 Inputs Required) ```bash /schliff Target: path/to/SKILL.md Goal: Make the skill trigger correctly for deployment scenarios ``` Defaults: Metric=composite_score, Verify=score-skill.py, Iterations=30. ## Core Loop (NEVER Pauses) ``` INPUT: Skill path + GOAL + PRIMARY METRIC + VERIFY method + time budget SETUP: Read ALL files → Analyze → Generate eval suite → Baseline (#0) LOOP (N iterations, continues until goal met or budget exhausted): Exp N: Review skill + results + git history → Pick ONE atomic change (based on gaps + history) → Edit SKILL.md or references → Commit: "schliff exp-N: [description]" → Run VERIFY, compute PRIMARY METRIC → Improved? Keep. Worse? Revert. Error? Fix or skip. → Append to history/ with diffs CONSTRAINT: Fixed iterations prevent infinite loops; autonomous mode = NO prompts between iterations, just continuous improvement. ``` ## When to Use - **Skill not triggering** → Run `/schliff` on trigger-accuracy metric - **Wrong/incomplete outputs** → Set goal, metric = binary eval pass rate - **Harden for edge cases** → Focus on edge-coverage metric - **Skill too verbose** → Optimize token-efficiency metric - **Do...