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 body delivers an unusually rigorous, well-sequenced writing workflow with genuine validation gates and concrete tool/file contracts, but it is severely overlong for a SKILL.md, repeats key rules multiple times, inlines content that belongs in its own reference files, and directs the agent to a templates/ bundle that is absent. Workflow clarity is excellent; token efficiency and file structure are the weak points.
Suggestions
Cut the body to an overview and move the memory rules, citation resource stack, writing-craft heuristics, and the full matrix row field enumeration into new files under references/, linked with purpose-stated 'read X when Y' pointers.
Deduplicate repeated rules — the AutoFigure-Edit caption sentence, the 30-50 references target, and the outline-folder sync procedure each appear two or more times; state each once and cross-reference.
Fix or remove the templates/ guidance: the skill claims ten venue template directories plus templates/README.md and templates/DEEPSCIENTIST_NOTES.md are bundled 'inside this skill', but none exist in the bundle, so the instruction currently dead-ends.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body runs ~1320 lines with substantial padding: the AutoFigure-Edit caption mandate appears twice (interaction discipline and Phase 6), the 30-50 verified-references target is stated twice (Phase 0 and Phase 5), outline-folder sync rules repeat across at least four sections, and long operating-contract material (milestone reporting cadence, QQ messaging, memory taxonomies) competes with writing craft. This fits 'Noticeably verbose; several unnecessary explanations or padded sections' — it is not level 1 (it does not explain concepts Claude already knows, and the guidance is domain-specific) but it is well below level 3's 'mostly efficient'. | 2 / 5 |
Actionability | Concrete, executable guidance dominates: exact tool calls ("artifact.submit_paper_outline(mode='candidate'|'select'|'revise', ...)"), specific paths (paper/outline/manifest.json, paper/evidence_ledger.json), a full matrix row schema, and complete JSON shapes for claim_evidence_map.json, figure_catalog.json, and submission_checklist.json. As an instruction-only skill, absence of code is acceptable per the scoring note, but some guidance stays abstract ('run a harsh self-review', 'adversarial mindset') without a concrete recipe, which keeps it at level 4 rather than the copy-paste-ready level 5. | 4 / 5 |
Workflow Clarity | Phases 0-9 are explicitly ordered with validation checkpoints throughout: a preconditions gate ('Before writing seriously, confirm:...'), a hard order for outline/ledger sync, a main-text drafting gate with row-status rules, a five-pass revision loop with recorded deltas, a submission gate, and exit criteria. Feedback loops for error recovery are explicit ('If any of those checks fails, stop drafting and repair the paper contract first', route back to experiment/analysis-campaign). This matches the level-5 anchor of clear sequencing, explicit validation, feedback loops, and checklists. | 5 / 5 |
Progressive Disclosure | References are real, one level deep, and clearly signalized with purpose ('read references/outline-evidence-contract-example.md before drafting further') — all seven referenced files exist in the bundle. But the SKILL.md is not an overview: ~1300 lines of material that belongs in reference files is inlined (memory rules, citation resource stack, drafting style heuristics, the full matrix field enumeration), and the claimed bundled templates directory ('Bundled templates live under templates/ inside this skill', ten template paths plus DEEPSCIENTIST_NOTES.md and README.md) does not exist in the bundle. This sits between level 2 ('content that clearly belongs in separate files is inlined') and level 3 ('some structure but could be better organized'); real, correct references and clear navigation lift it to 3. | 3 / 5 |
Total | 14 / 20 Passed |