Content
82%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.
A dense, highly actionable skill body with executable commands, precise data contracts, and a well-structured reference routing table. Weaknesses are limited to readability of the overloaded workflow steps, implicit (rather than explicit) failure-recovery loops around the quality gate, and a few unrouted bundle files.
Suggestions
Split the overloaded workflow steps (especially 1, 3 and 6) into shorter sub-bullets so each rule is individually scannable.
Add an explicit quality-gate failure path (e.g. 'if the gate fails: fix coverage, re-run the pilot, re-gate before scaling') to close the feedback loop.
Either route references/pitfalls.md and the unrouted scripts from the task table, or remove them from the bundle so every file is discoverable.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and rule-per-sentence with no padding and no explanation of concepts Claude already knows ('lookalikes are not wins', 'Reviews are not calls, openings are not hires'), but workflow steps 1-3 and 6 are overloaded multi-clause paragraphs that could be split or trimmed, matching 'minor instances of over-explanation that could be tightened' rather than the lean top anchor. | 4 / 5 |
Actionability | Fully executable guidance throughout: copy-paste-ready 'python3 scripts/analyze_signals_v2.py' invocations with realistic flags, an exact CSV column contract ('domain,status,website,jobs; optional account_id,parent_id,split,known_at,scored_at'), concrete JSON source-adapter shapes, and config schema examples covering the common discovery and validation cases. | 5 / 5 |
Workflow Clarity | The 9-step workflow has a clear sequence and explicit checkpoints (pilot one or two rows then pass the quality gate, freeze artifacts before validation, blocked-state handling for missing prerequisites), but what to do when the quality gate fails is only implicit ('Imputation or dropping missingness flags can still encode collection bias') rather than an explicit fix-and-revalidate loop, keeping it below the top anchor. | 4 / 5 |
Progressive Disclosure | Excellent one-level-deep structure: a task-indexed routing table pointing to 16 real reference files, all of which exist in references/, plus routed scripts. Minor gaps: references/pitfalls.md is an orphan never linked from the body, and several bundle scripts (scoring_contract.ts, replay-score.play.ts, dedupe_utils.py, technology_evidence.py, find_contacts_v2.py) have no body-level navigation. | 4 / 5 |
Total | 17 / 20 Passed |