Content
70%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 genuinely expert, actionable paper-writing guidance with excellent workflow checklists, validation checkpoints, and executable code. Its weaknesses are redundancy (citation rules and conference details repeated across sections, philosophy tables duplicated with the references) and navigation defects from dangling templates/ references and inlined deep-dive content.
Suggestions
Consolidate the citation-hallucination rule into a single section — it is currently stated in the opening CRITICAL section, the Golden Rule block, the summary table, and the closing line; keep one canonical statement plus the summary table.
Move the full 7-principles table, Ethan Perez micro-tips, and Lipton word-choice lists entirely into references/writing-guide.md, keeping only one-line summaries with links inline to cut the ~940-line body.
Fix or remove the dangling templates/ references (templates/README.md, templates/neurips2025/, etc.) — the directory does not exist in the bundle, so the template-setup workflow is not executable as shipped.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is substantive domain guidance rather than generic-concept padding, but it could clearly be tightened: the never-hallucinate-citations rule is stated three to four times (opening CRITICAL section, Golden Rule, summary table, closing line), conference requirements appear in three overlapping tables, and the full 7-principles and micro-tips tables are inlined while also living in writing-guide.md. Dated venue details ("NeurIPS 2025", "ICML 2026" page limits) add time-sensitive bulk outside any deprecated section. | 3 / 5 |
Actionability | Mostly executable guidance: runnable Python for Semantic Scholar search and DOI-to-BibTeX, bash repo-exploration and template commands, LaTeX snippets, and MCP install commands. Minor gaps keep it from the top anchor — "Verify the claim actually appears in the paper" gives no executable method, and the template copy commands target a templates/ directory that is absent from the bundle. | 4 / 5 |
Workflow Clarity | Four workflows are laid out as copyable checklists with explicit validation checkpoints (compile the unmodified template first, verify papers in 2+ sources, verify claims before citing) and feedback loops for error recovery (placeholder protocol plus informing the scientist, fixing template compile issues before editing). This matches the top anchor: clear sequences, validation steps, and checklists. | 5 / 5 |
Progressive Disclosure | The eight references/ files are real, one level deep, and well signaled both inline and in a summary table — but the score is dragged down because deep-dive philosophy content (7-principles table, micro-tips, word-choice lists) is duplicated inline instead of being split into writing-guide.md, and repeated references to templates/ and templates/README.md dangle because no templates directory exists in the bundle. | 3 / 5 |
Total | 15 / 20 Passed |