Content
82%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 highly actionable, well-sequenced planning workflow that gives Claude everything needed to execute: exact filenames, fill-in templates, venue-specific rules, a reviewer-model cross-check, and a defined output artifact. The few weaknesses are redundancy in venue/page-budget rules, a single-pass reviewer loop, and reliance on external reference files that are not bundled with the skill.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is almost entirely operational (templates, tables, exact filenames, tool configs) and assumes Claude's competence — no concept explanations. Not 5 because the venue page-count rules and IEEE inclusion caveat are stated twice (Constants and Key Rules), and the front-loading guidance repeats across the overlay section, Step 2, and Key Rules. | 4 / 5 |
Actionability | Guidance is fully concrete and executable: exact input filenames, a fill-in Claims-Evidence Matrix, per-section planning templates with page budgets, a figure-plan table, a copy-paste MCP reviewer invocation (model, config, full prompt), and a complete PAPER_PLAN.md output template. Templates are fill-in-the-blank by design, appropriate for a generative planning skill, and cover the common paper types (empirical, theory, method). | 5 / 5 |
Workflow Clarity | The 7-step workflow is clearly sequenced with branching at Step 1 (use CLAIMS_FROM_RESULTS.md if present, else extract from scratch; ask the user if no inputs exist) and explicit constraints ("total page budget MUST sum to MAX_PAGES", section-count match). Not 5 because the Step 6 reviewer feedback is a single pass — "Apply feedback before finalizing" — with no re-review iteration or threshold for when the outline is good enough. | 4 / 5 |
Progressive Disclosure | Clear section headers organize the body, and external references (writing-guide.md, checklists.md) are one level deep with explicit load-only-when-needed signaling. Not 5 because no bundle files ship with this skill — the referenced ../ml-paper-writing files live outside it and cannot be verified — and the dense per-section templates are all inlined in a ~275-line file rather than split out. | 4 / 5 |
Total | 17 / 20 Passed |