Content
48%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 shows strong workflow discipline — numbered sequences, explicit verification gates, and a self-verification checklist — and its bundle references are real and well-integrated. It is dragged down by substantial redundancy (triple-listed paper lanes, duplicated lifecycle rules) and by keeping ~130 lines of paper-search detail inline in an already long SKILL.md instead of a reference file.
Suggestions
Keep one canonical list of paper/preprint lanes (e.g. in a references/paper-search-lanes.md) and reference it from Paper Radar, Multi-Source Mesh, and Active Research instead of repeating it three times.
Move the arXiv route, Paper Radar Pattern, and Multi-Source Mesh sections (~130 lines) into a reference file, leaving a short pointer plus the script command in SKILL.md.
Bundle or inline the essential steps of the managed intake (the content of .local/material-lifecycle/README.md and the usage of material_wave_ranking.py) so the core intake workflow is executable without files absent from the skill.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is noticeably padded: the venue/preprint lane list (NeurIPS, ICLR, ICML, ACL, EMNLP, AAAA, bioRxiv, medRxiv, ChemRxiv) appears three times (Paper Radar Pattern, Multi-Source Paper Exploration Mesh, Active Research Workflow), intake/lifecycle governance rules are duplicated between the ~50-line Boundary section and the Intake Workflow, and 请你读/精读 routing is stated four times (mode table, directive convention, step 0, detail-preservation preamble). This matches 'several unnecessary explanations or padded sections' rather than the mostly-efficient anchor 3. | 2 / 5 |
Actionability | There is genuinely concrete guidance — the copy-paste script invocation with flags, the numbered arXiv URL resolution order with exact URL patterns, the per-platform reader mapping (微信公众号 -> wechat-article-reader), and the closed disposition enum (top_window, ranked_backlog, no_change). However, the core managed-intake steps defer to files not in the bundle ('.local/material-lifecycle/README.md', 'note-system/materials/material_wave_ranking.py'), leaving key executable details missing, which fits anchor 3 rather than 4. | 3 / 5 |
Workflow Clarity | Multi-step processes are clearly sequenced (Intake steps 0-6, Active Research 1-7, arXiv resolution 1-6) with explicit validation gates ('Verify an intake adds exactly one record... then create the separate... ranking revision'; 'Verify the disposition... authority readback... before reporting completion') and a dedicated Self-Verification checklist including a user-highlighted point audit. It falls short of 5 because the complete intake flow must be assembled from three interleaved sections (Boundary, Intake Workflow, Self-Verification) rather than one coherent sequence, and error-recovery loops are implicit outside the Unread/PDF-fallback paths. | 4 / 5 |
Progressive Disclosure | Against the actual bundle, the two references (references/paper-reading-protocol.md, references/decision-driven-scouting.md) are real, well-signaled markdown links, and are one level deep (no further .md references inside them); the script is real and linked with a usage example. But the 361-line body inlines ~130 lines of paper-search machinery (arXiv route, Paper Radar Pattern, Multi-Source Paper Exploration Mesh) that clearly belongs in a bundled reference file, so 'content that should be separate is inline' holds more strongly than the 'minor organization gaps' of anchor 4. | 3 / 5 |
Total | 12 / 20 Passed |