Content
77%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 high-quality, highly operational body: executable commands dominate, ownership rules are explicit, and risky batch/destructive paths all carry validation or gating. The weaknesses are structural and editorial — heavy inlined reference material with no bundle split, and war-story meta-commentary about past skill versions that costs tokens without guiding execution.
Suggestions
Move the paper-page schema, the query-pack budget table, and the per-skill hook contracts (Hooks 1-4) into one-level-deep reference files (e.g. references/paper-schema.md, references/query-pack.md, references/integration-hooks.md) and link them from the body, keeping SKILL.md as an overview.
Cut the past-version drift anecdotes (the 'earlier versions of this skill described a prose-only init' paragraph and the 'left a real user's research-wiki/ empty for a week' aside) — they narrate skill history rather than instruct, and the recovery path in the resolution block already conveys the fix.
Close the actionability gaps by showing the helper command forms for `update` and `stats` (as is done for ingest/sync/add_edge), and drop the duplicated Karpathy acknowledgement in favor of the single Overview mention.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense, non-obvious operational detail (resolution chain, exact schemas, budgets, ownership rules) with no padding explaining things Claude already knows. But several passages could be trimmed: two past-version drift anecdotes ("Earlier versions of this skill described a prose-only init that omitted query_pack.md — that drifted...", "the failure mode that left a real user's research-wiki/ empty for a week") and a duplicated Karpathy attribution (Overview plus Acknowledgements). | 4 / 5 |
Actionability | Copy-paste-ready bash for the helper-resolution chain, init, ingest (arXiv and venue variants), sync, add_edge, upsert_idea, add_experiment, and add_claim, plus the exact emitted page schema and budget table. Minor gaps keep it off the top anchor: the `update` subcommand shows slash-command examples without the underlying helper invocation, and `stats` shows only an output sample with no command. | 4 / 5 |
Workflow Clarity | Explicit validation and feedback loops throughout: the helper-resolution block hard-fails with a numbered 4-option recovery path; Hook 3 gates supports/invalidates edges on EXP_NODE_OK (node born before edges); batch `sync` validates per-id with silent dedup; UTF-8 conversion mandates a backup first; and `lint` provides a 6-point failure-mode checklist. Sequencing is unambiguous with a mandatory pre-step (helper resolution) before every subcommand. | 5 / 5 |
Progressive Disclosure | The single SKILL.md is well-sectioned with clear headers, but there is no bundle at all (no references/, scripts/, or assets/), and substantial content that belongs in one-level-deep reference files is inlined: the full paper-page schema, the query-pack budget table, and three per-skill hook contracts. The only file references (../shared-references/capture-antipatterns.md, ../shared-references/integration-contract.md) point outside this skill's bundle, matching the 'content that should be separate is inline' anchor. | 3 / 5 |
Total | 16 / 20 Passed |