Content
100%Weight 40%Scale 1-5Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
The body is a dense, executable guide with a well-sequenced workflow, strong validation gates, and clean one-level-deep reference navigation. It respects the token budget while remaining highly actionable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean, imperative prose throughout; no padding or re-explanation of concepts Claude already knows — every line is actionable guidance about preserving the operating model, preflight, cognition, and round discipline. | 5 / 5 |
Actionability | Concrete wrapper commands are named (scripts/evolve-brief normalize, scripts/evolve-cognition init/add, evolve-db record/sample/best/stats) alongside an explicit 10-step round loop and specific configuration knobs (sampling.algorithm, island feature semantics, timeout handling). | 5 / 5 |
Workflow Clarity | A clear sequenced round loop with explicit validation/confirmation checkpoints (keep approval.confirmed=false until user confirms; refuse to mutate/evaluate before confirmation; mandatory per-round database sample) plus feedback loops (record lesson, check best snapshot, decide whether another round is justified). | 5 / 5 |
Progressive Disclosure | The body is an overview that points to one-level-deep reference files (operating_model.md, preflight.md, run_spec.md, toolbelt.md, architecture.md) in a clearly signaled 'Read when needed' section; all referenced files exist and the split is appropriate. | 5 / 5 |
Total | 20 / 20 Passed |