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 thorough, well-sequenced writing workflow with real validation checkpoints, feedback loops, and a final checklist. Its main weaknesses are significant redundancy (the calibration/limitations doctrine is stated three times in different forms), a pseudocode-only bib-cleaning step, and a broken disclosure structure where referenced files (`templates/`, `../shared-references/`) are not present in the bundle while large reference-grade content is inlined.
Suggestions
Consolidate the duplicated calibration/limitations guidance: merge the 12-rule 'CONFIDENT PROSE, HONEST LIMITS' block with the overlapping Step 6 review prompt text and the 'Calibrate, don't hedge' / 'Launch, not progress report' Key Rules bullets into a single authoritative section (or a reference file), and remove the repeated 'After drafting all sections' lead-ins.
Replace the comment-only 'Automated bib cleaning' Python pattern with runnable code that greps `\citep{}`/`\citet{}`/`\cite{}` keys and filters the .bib file, since this is the one non-executable step in an otherwise copy-ready workflow.
Fix the disclosure structure: either ship the referenced `templates/` and `../shared-references/*.md` files in the bundle or drop/inline those references, and move the five-pass audit detail and prose-style rules into a single bundled reference file linked from a short overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~480-line body contains several noticeably padded or duplicated sections: the 12-rule "CONFIDENT PROSE, HONEST LIMITS" block restates the same calibration/limitations guidance that already appears in the Step 6 review prompt, the §5 Conclusion guideline, and again as the "Calibrate, don't hedge" / "Launch, not progress report" Key Rules bullets; Step 5 states "After drafting all sections" twice with overlapping sentence-clarity rules in Pass 3 and the later bulleted list. This exceeds anchor 3 ("some unnecessary explanation or could be tightened") and fits anchor 2 ("several unnecessary explanations or padded sections"), though it is not the pervasive explanation-of-basics that would merit a 1. | 2 / 5 |
Actionability | Most guidance is executable: copy-ready `curl` commands for the DBLP/CrossRef fetch chain, concrete LaTeX documentclass/author-block snippets, an explicit project file tree, per-section page targets, and a fully specified MCP review call with job polling. The main gap is the "Automated bib cleaning" Python pattern, which is only comment-style pseudocode ("use this Python pattern" with numbered comments, no runnable code), so it fits anchor 4 ("concrete code or commands with minor gaps") rather than fully copy-paste-ready anchor 5. | 4 / 5 |
Workflow Clarity | Steps 0–8 are clearly sequenced with explicit validation checkpoints and feedback loops: backup before overwriting the destructive `paper/` step, stale-file cleanup rules, the three-step bibliography fallback chain ending in a `[VERIFY]` marker, cross-review with CRITICAL/MAJOR/MINOR fix triage and poll-until-done instructions, a restatement audit, and a final 14-item checklist. This matches anchor 5 ("clear sequence with explicit validation steps; feedback loops for error recovery; checklists"). | 5 / 5 |
Progressive Disclosure | References to `../shared-references/writing-principles.md`, `venue-checklists.md`, and `citation-discipline.md` are well-signaled with when-to-read guidance, but none of those files exist in the provided bundle, and "The skill includes conference templates in `templates/`" points to a directory that is also absent. Meanwhile the 65-line 12-rule prose block and the five-pass audit detail are inlined material that clearly belongs in a reference file. That combination — present-but-unverifiable references plus inlineable bulk — fits anchor 3 ("some structure but could be better organized"), above anchor 2 only because the reference signals themselves are explicit and one level deep. | 3 / 5 |
Total | 14 / 20 Passed |