Content
70%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 highly actionable, validation-rich skill body whose helper commands, schemas, and error-recovery gates are exemplary. Its weaknesses are token overhead from design-rationale storytelling and a monolithic structure that inlines material (page schema, integration hooks, query-pack budgets) that should live in bundled reference files.
Suggestions
Move the per-entity page schemas, the query-pack budget table, and the Hooks 1–4 integration details into bundled reference files (e.g. references/page-schemas.md, references/integration-hooks.md) and keep SKILL.md as a subcommand overview with one-level links, which would raise both conciseness and progressive disclosure.
Add the concrete helper invocations for the update, lint, and stats subcommands (e.g. python3 "$WIKI_SCRIPT" lint research-wiki/) so all seven subcommands are copy-paste ready, and convert the pseudocode in Hooks 1 and 3 (update_idea, rebuild query_pack) into actual commands.
Trim narrative asides that add tokens without instruction value: the init drift-history parenthetical, the duplicated Karpathy acknowledgement, and the rationale prose around the capture-hygiene and helper-resolution sections.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense and operational with no explanation of concepts Claude already knows, but carries removable padding: a drift-history parenthetical about a past failure, the Karpathy attribution duplicated in Overview and Acknowledgements, and prose rationale around capture hygiene — matching anchor 3 ('could be tightened') rather than 4. | 3 / 5 |
Actionability | Copy-paste-ready bash for helper resolution, init, ingest_paper (three variants), sync, add_edge, upsert_idea, add_experiment, and add_claim, plus a verbatim page schema. Minor gaps hold it at anchor 4: the update, lint, and stats subcommands show no concrete helper invocation, and Hooks 1/3 mix in pseudocode like 'update_idea(active_idea_id, outcome=verdict)'. | 4 / 5 |
Workflow Clarity | Explicit validation and recovery loops throughout: the resolution chain exits with four numbered remediation options, capture-filter flags trigger rewrite-or-drop, EXP_NODE_OK gates edge writes against a missing node, sync dedups per-id, and lint is a 6-item checklist producing a fix report — matching anchor 5 (explicit validation, feedback loops, checklists) rather than 4. | 5 / 5 |
Progressive Disclosure | Well-sectioned headers and clearly signaled one-level references exist, but the ~460-line body inlines the full paper-page schema, the query-pack budget table, and ~100 lines of integration hooks for other skills that belong in separate reference files; no bundle files exist, and the only references (../shared-references/*.md) sit outside the bundle and do not resolve — matching anchor 3. | 3 / 5 |
Total | 15 / 20 Passed |