Content
62%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 skill body is highly actionable and unusually disciplined about validation — verification gates, coverage receipts, graceful degradation, and checklists are all explicit. However, it is severely bloated by seven verbatim copies of the helper-resolution block and quadruple-stated source-selection rules, and it inlines reference-level policy content while linking to ../shared-references/ files that are absent from the bundle.
Suggestions
Factor the 9-line $ARIS_REPO/$FETCHER resolver into one canonical block (or a single referenced helper) and have each source section say only 'resolve $X_FETCHER per the canonical chain in §Resolver' — this removes ~60 lines of duplication and eliminates the 'both lines must stay in sync' maintenance hazard.
State the source-selection rule (default tier vs. opt-in fetchers) exactly once — in the Source Selection section — and delete the restatements in the Constants block, the Source Table rows, and the D2 finalization block; replace the fragile line-number pointer 'L42-43 / L51-63 of this SKILL' with a section anchor.
Move the D2 policy prose, the per-source de-duplication rules, and the detailed resolver contract into a references/ file inside the bundle, and either vendor the ../shared-references/*.md files into the bundle or remove/repair those dead links so every referenced path resolves.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~770-line body duplicates the 9-line $ARIS_REPO resolver bash block verbatim seven times (arXiv, S2, DeepXiv, Exa, OpenAlex, verify, wiki sections), and restates the source-selection rule ("semantic-scholar, deepxiv, exa, gemini, and openalex are excluded from all") in at least four places, including meta-commentary like "both lines must stay in sync" and "matching the existing convention at L42-43 / L51-63 of this SKILL". That is several padded/redundant sections — the 2 anchor ('noticeably verbose; several unnecessary explanations or padded sections') — though it never teaches concepts Claude already knows, so it stays above 1. | 2 / 5 |
Actionability | Quotes: complete executable bash blocks ("python3 \"$ARXIV_FETCHER\" search \"QUERY\" --max 10"), a copy-paste Gemini MCP prompt with full per-paper field requirements, a complete fallback python3 heredoc for verified_papers.json, and an exact output table format. Minor gaps keep it below 5: 'QUERY'/'ARXIV_ID' placeholders are never tied to a concrete substitution step, and Step 0a instructs "Try the most common patterns and adapt" for unknown MCP tool names. | 4 / 5 |
Workflow Clarity | Quotes: a fully ordered Step 0a→6 pipeline; "Verify Candidate Papers (anti-hallucination, mandatory)" with a 3-layer verification chain, degraded-output fallback, and an explicit BLOCK path ("Status: BLOCKED. Install python3 or restore the helper"); the D2 empty-aggregate error gate ("if zero sources contributed, surface a D2 empty-aggregate error and stop"); a coverage receipt reconciling covered vs. uncovered IDs; and a per-item checklist in Step 6 with retry/fallback on network errors. This is a clear sequence with explicit validation, error-recovery feedback loops, and checklists — the 5 anchor; the batch fan-out in Step 2 does have its coverage-receipt validation, so no cap applies. | 5 / 5 |
Progressive Disclosure | There are no bundle files (no references/, scripts/, or assets/), and the body's five markdown links point to ../shared-references/*.md which do not exist on disk — dead reference targets. Structure exists (clear ## sections, a source table, well-signaled links), but ~400 lines of content that clearly belongs in separate reference files (the repeated resolver, the D2 policy prose, the per-source de-duplication rules) are inlined in SKILL.md. This matches the 3 anchor ('some structure; references present but content that should be separate is inline') rather than 2, since sections and links are clearly signaled rather than buried. | 3 / 5 |
Total | 14 / 20 Passed |