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.
This is a highly actionable, exceptionally well-sequenced workflow with strong validation and feedback loops, but it is bloated: roughly 690 lines with significant redundancy (rules restated, venue rules repeated three times) and background material Claude already knows. It also functions as a monolith, inlining scripts and templates that belong in bundle files while pointing at external references that do not resolve within the bundle.
Suggestions
Deduplicate the Key Rules section against the CONFIDENT PROSE, HONEST LIMITS block and state the venue citation-style and page-limit rules once (in Constants), removing the three repetitions of each; this alone should cut substantial length.
Move the inline 70-line BibTeX hygiene validation script to scripts/validate_bib.py (invoked with a one-line command) and the venue LaTeX setups to templates/ files, keeping SKILL.md an overview with well-signaled one-level-deep references.
Trim Pass 1's cluttered-phrase replacement table and AI-isms list to a single sentence of intent — Claude already knows 'due to the fact that' means 'because' and that 'delge/pivotal/landscape' are filler — keeping only the non-obvious Banana Rule and venue-specific guidance.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | There are several padded or redundant sections beyond 'some unnecessary explanation': the 12-rule CONFIDENT PROSE block is restated in Key Rules ("rules 9-12 above"), venue citation-style and page-limit rules appear three times (Constants, Step 3, Key Rules), related-work >=1 page appears three times, and Pass 1 teaches Claude things it already knows (a "Due to the fact that -> Because" table, "In order to -> To", and an AI-isms list of delve/pivotal/landscape). It is not a 1 because most content is genuinely task-specific and non-obvious (venue templates, DBLP/CrossRef chain, IEEE-vs-ML rules). | 2 / 5 |
Actionability | The body is overwhelmingly copy-paste ready: executable curl commands for DBLP/CrossRef, a fully runnable 70-line Python validation script, concrete LaTeX template snippets, an exact project tree, and a complete reviewer MCP prompt payload. The small gap is the 'Automated bib cleaning' block, which is a comment skeleton rather than executable code ("# 1. Grep all \citep{...} ... # 4. Write the filtered bib"), so it falls just short of the fully-executable 5 anchor. | 4 / 5 |
Workflow Clarity | Steps 0-8 are clearly sequenced with explicit validation checkpoints and feedback loops: backup before overwrite, stale-file cleanup, a mandatory bib hygiene validation run immediately after writing references.bib with explicit DEAD/VERIFY/MISMATCH remediation paths, a theory restatement audit before Step 4, and a final 16-item checklist. This matches the anchor for clear sequence, explicit validation, error-recovery loops, and checklists; even destructive operations (overwriting paper/) have a backup gate. | 5 / 5 |
Progressive Disclosure | References are clearly signaled and purpose-scoped ("Read ../shared-references/writing-principles.md before drafting the Abstract... only when prose feels generic"), which is good, but none of the referenced files (../shared-references/*.md, templates/, tools/extract_paper_style.py) exist in the provided bundle, and heavy content that clearly belongs in separate files is inlined — the 70-line validation script, the venue template listings, and the 12-rule prose block inside a 690-line monolithic SKILL.md. This sits between the 2 anchor (content that belongs in separate files is inlined) and the 4 anchor (most content appropriately placed), with structure present but organization improvable. | 3 / 5 |
Total | 14 / 20 Passed |