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 is a well-engineered, highly actionable pipeline: sequenced phases with hard checkpoints, executable scripts and MCP prompts, a state-resume mechanism, and a feedback-driven revision loop. Its weaknesses are token efficiency (duplicated pipeline diagrams and repeated agency criteria) and a monolithic single-file layout that inlines per-agency specifications and review-bundle templates that belong in reference files.
Suggestions
Move the per-agency Grant Type Specifications and the Codex review-bundle prompt templates into reference files (e.g., references/grant-types.md and references/review-bundles.md), keeping only a summary table and the agency list in SKILL.md — this would address both progressive disclosure and the ~700-line token cost.
Eliminate duplicated content: the funding/publish track is drawn twice (the Overview ASCII diagram and the "Composing with Other Skills" section) — keep one; and the Grant-Specific Drafting Guidelines repeat criteria already tabulated in Grant Type Specifications, so merge them.
Tighten the remaining vague spots for full actionability: replace "[INSERT GRANT-TYPE-SPECIFIC CRITERIA]" with a pointer to the exact criteria table rows to copy, and specify how to generate the SVG/matplotlib figures (or route them through the /paper-illustration invocation unconditionally).
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~700-line body avoids explaining known concepts but has clear slack: the funding-vs-publish track is diagrammed twice (Overview ASCII art and again under "Composing with Other Skills"), the Grant Type Specifications tables and the Grant-Specific Drafting Guidelines repeat criteria, and the opening reviewer-routing block is dense. Fits 'mostly efficient but includes some unnecessary explanation or could be tightened' rather than the minor-trimming of a 4. | 3 / 5 |
Actionability | Highly concrete throughout: a copy-paste-ready bash resolver for the style helper, an exact GRANT_STATE.json schema, complete MCP prompt templates with threadId reuse, per-agency section tables, and a gap-statement template. Minor gaps keep it below 5: the review bundle requires the reader to assemble "[INSERT GRANT-TYPE-SPECIFIC CRITERIA]" and figure guidance ("generate clean SVG or matplotlib directly via code") is vague. | 4 / 5 |
Workflow Clarity | Phases 0-5 are explicitly sequenced with ⛔ STOP checkpoints and user-response options, state persistence has resume/staleness rules, revisions are severity-tiered (CRITICAL/MAJOR/MINOR) with re-review rounds, and a final checklist validates the output. This matches the top anchor: explicit validation steps, feedback loops for error recovery, and a checklist for a complex process. | 5 / 5 |
Progressive Disclosure | No bundle files exist (references/, scripts/, assets/ all absent), so everything lives in one ~700-line body. The per-agency specification tables and full Codex bundle templates are inlined content that clearly belongs in separate reference files, while the external pointers (../shared-references/*.md, tools/extract_paper_style.py) are clearly signaled but the bulk detail is not split out. Fits 'some structure but could be better organized; content that should be separate is inline'. | 3 / 5 |
Total | 15 / 20 Passed |