Content
75%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 is a well-structured, actionable planning workflow with a clear 7-step sequence, a reviewer feedback loop, and concrete templates plus an executable Codex MCP call. It is token-efficient for a complex skill, though the repeated per-section template fields and inlined templates leave minor conciseness and progressive-disclosure trim opportunities.
Suggestions
Collapse the repeated per-section template boilerplate (e.g. 'Estimated length', 'Self-contained check', 'Front-loading check') into a single shared field legend to reduce token overhead across the §0–§5 templates.
Move the detailed section-by-section template block into a reference file (e.g. references/section-templates.md) and keep SKILL.md as an overview pointing to it, strengthening progressive disclosure.
Tighten Step 6 by making the review outcome an explicit pass/revise checkpoint (e.g. 'if any score < 7, revise and re-review') so the feedback loop has a concrete validation gate rather than a soft 'apply feedback before finalizing'.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dense instructional reference material (Claims-Evidence Matrix, section templates, figure/citation plans) that mostly earns its tokens without explaining concepts Claude already knows, matching 'Efficient; minor instances of over-explanation that could be trimmed'. Not a 5 because repeated fields like 'Estimated length' and self-checks ('Front-loading check', 'Self-contained check') on every section template, plus the trailing Acknowledgements, could be trimmed; not a 3 because there is no padded conceptual explanation of the kind the lower anchor penalizes. | 4 / 5 |
Actionability | Quotes concrete executable guidance: the Codex MCP call block with model 'gpt-5.4' and config '{"model_reasoning_effort": "xhigh"}', plus fill-in templates for the matrix, figure plan table, and citation plan, matching 'Mostly executable guidance; concrete code or commands with minor gaps'. Not a 5 because the section templates are scaffolds with placeholders rather than copy-paste-ready output covering common cases; not a 3 because the MCP call and table formats are concrete and usable, not pseudocode. | 4 / 5 |
Workflow Clarity | Quotes a clear 7-step sequence (Extract Claims → Determine Structure → Section Planning → Figure Plan → Citation Scaffolding → Cross-Review → Output) with a feedback loop in Step 6 ('Apply feedback before finalizing') and a Key Rules checklist, matching 'Clear sequence with most checkpoints present; minor validation gaps'. Not a 5 because the Step 6 review is a soft review-then-apply step rather than an explicit pass/fail validation with error-recovery retry; not a 3 because steps are well-sequenced with a real feedback loop. This is a planning (non-destructive, non-batch) skill so the destructive-cap does not apply. | 4 / 5 |
Progressive Disclosure | Quotes well-signaled one-level-deep references ('Read ../shared-references/writing-principles.md', 'Read ../shared-references/venue-checklists.md') with an explicit instruction to load only when needed, plus clear section headers, matching 'Good structure; most content is appropriately placed; references mostly clear'. Not a 5 because all the per-section templates are inlined in SKILL.md rather than split into a reference file, and no bundle files are present in the reviewed package to confirm the shared-references targets exist; not a 3 because the references that do exist are clearly signaled and exactly one level deep. | 4 / 5 |
Total | 16 / 20 Passed |