How alchaincyf/darwin-skill works
Darwin Skill is an agent skill for iteratively improving other SKILL.md files. Its principal workflow evaluates a baseline, edits one dimension, independently validates the change, keeps improvements or reverts regressions, and pauses for human review. README documentation supplies the workflow and capabilities; the only sampled source is a peripheral screenshot script, so stage-level implementation details are attributed to SKILL.md without claiming its contents were inspected.
Main entry points: SKILL.md Darwin Skill · result-card.html Result Card · test-prompts.json Test Prompts
Components 14
Skill Workflow
- Darwin Skill
SKILL.md - Target Skill
- Human Checkpoint
SKILL.md - Result Card
templates/result-card.html
Evaluation
- Baseline Review
SKILL.md - Nine-Dimension Rubric
SKILL.md - Structural Scoring
SKILL.md - Effectiveness Testing
SKILL.md
Validation
- Test Prompts
test-prompts.json - Independent Reviewers
SKILL.md - Regression Review
SKILL.md
Iteration
- Single-Dimension Edit
SKILL.md - Keep or Revert
SKILL.md
Other components
- User
Connections 14
- User → Darwin Skill: requests optimization
README.md - Darwin Skill → Target Skill: optimizes
README.md - Darwin Skill → Baseline Review: starts with
README.md - Baseline Review → Nine-Dimension Rubric: scores against
README.md - Baseline Review → Structural Scoring: assesses structure
README.md - Baseline Review → Effectiveness Testing: assesses results
README.md - Darwin Skill → Single-Dimension Edit: proposes change
README.md - Single-Dimension Edit → Independent Reviewers: requests validation
README.md - Independent Reviewers → Keep or Revert: provides scores
README.md - Keep or Revert → Target Skill: keeps or reverts
README.md - Darwin Skill → Human Checkpoint: pauses for approval
README.md - Darwin Skill → Test Prompts: uses for testing
README.md - Darwin Skill → Regression Review: runs regression tests
README.md - Darwin Skill → Result Card: presents results