Content
56%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 strong, expert-voiced instruction body with unusually concrete formatting and verification rules and a logical workflow with real checkpoints. Its weaknesses are the missing kernel.py file that every executable instruction depends on, and cross-section redundancy in the prose-style guidance that inflates token cost without adding instruction.
Suggestions
Ship kernel.py in the skill bundle (or inline the critical helper signatures) — the body's only executable code (`exec(open("<this skill's directory>/kernel.py").read())`) fails as bundled, since no kernel.py exists in the skill directory.
State the kernel.py helper list once (in Setup) and drop the restatement in the "Put the answer" section, and merge the overlapping prose-style rules in "Making the prose carry its weight" and "Put the answer in the answer" to cut redundant tokens.
Add one short example invocation per key helper (e.g. `search_openalex("transformer scaling laws")`, `verify_dois(["10.xxxx/xxxx"])`) so the actionability is copy-paste ready rather than name-only.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The prose is dense and opinionated with little explanation of concepts Claude already knows, but it is noticeably redundant: the kernel.py helper list is repeated in the "Put the answer" section, and the process-narration / open-on-substance rules span two overlapping sections ("Making the prose carry its weight" and "Put the answer in the answer"), so it fits anchor 3 rather than anchor 4's minor trimming. | 3 / 5 |
Actionability | Guidance is concrete (exact exec line for kernel.py, named env vars, exact citation markdown format with %28/%29 encoding rules, style_pass single-pass usage), but the core executable dependency — kernel.py — is referenced throughout yet absent from the bundle, leaving the executable path incomplete per anchor 3's "missing key details" rather than anchor 4's mostly-executable bar. | 3 / 5 |
Workflow Clarity | The sections sequence a coherent workflow (setup → read the request → retrieval sweep → citation-graph expansion → retraction checks → synthesis → style pass → save) with validation checkpoints such as verify_dois, notice inspection for surprising findings, and style_pass before saving with an explicit fix-in-one-pass rule. Minor gaps — no numbered overview and only implied handling of a missing OPENALEX_API_KEY — keep it at anchor 4 rather than 5. | 4 / 5 |
Progressive Disclosure | Section headers are clear and references are one level deep, but the single referenced bundle file (kernel.py) does not exist in the bundle, and dense inline style rules (citation link formatting, DOI URL-encoding, heading-length rules) read like content that belongs in a separate reference file — fitting anchor 3's structure-present-but-imperfect placement rather than anchor 4. | 3 / 5 |
Total | 13 / 20 Passed |