Content
77%Reviews the quality of instructions and guidance provided to agents. Good implementation is clear, handles edge cases, and produces reliable results.
A highly actionable, clearly sequenced grant-writing workflow with strong checkpoints, state persistence, and feedback loops. Its main weakness is conciseness and progressive disclosure: the file is a long monolith with large reference-style tables and prompt templates kept inline rather than factored into bundled reference files.
Suggestions
Move the per-agency specification tables and grant-specific drafting guidelines into a bundled reference file (e.g. references/GRANT_TYPES.md) and link to it from the body, keeping SKILL.md as a lean overview — this would improve both conciseness and progressive_disclosure.
Extract the reusable Codex review prompt templates (Round 1 / Round 2+) into references/REVIEW_PROMPTS.md or a script, replacing the inline blocks with a short pointer, to cut significant inline tokens.
Tighten the duplicated pipeline/track diagrams (the Overview ASCII diagram, the Constants note, and the 'Composing with Other Skills' section restate the same branching pipeline) into a single canonical diagram to reduce redundancy.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is densely informative and largely free of beginner-concept padding, but at ~600 lines it is long: the per-agency cultural-norms tables, the parallel Codex round-1/round-2 prompt blocks, and the duplicated pipeline diagrams could be tightened or pushed to a reference, so it is 'mostly efficient but could be tightened' rather than fully lean. | 2 / 3 |
Actionability | It gives concrete executable guidance throughout: ready-to-paste Codex MCP prompts with threadId/config, exact sub-skill invocations (`/aris-research-lit "$ARGUMENTS"`), a fillable claims-aims-evidence matrix, file-output trees, and a copy-ready final-checklist — copy-paste ready rather than pseudocode. | 3 / 3 |
Workflow Clarity | Five phases are explicitly sequenced (Phase 0→5) with 🚦/⛔ checkpoints, state persistence (GRANT_STATE.json with resume rules), feedback loops (review→CRITICAL/MAJOR fixes→re-review, MAX_REVIEW_ROUNDS), and a final verification checklist, matching the 'clear sequence with explicit validation steps and feedback loops' anchor. | 3 / 3 |
Progressive Disclosure | There are no references/scripts/assets bundles and the skill is monolithic: large reference-worthy tables (per-agency specs, cultural norms, parameter pass-through) and the verbose review-prompt templates live inline rather than being split into one-level-deep reference files, fitting 'content that should be separate is inline' more than the well-signaled multi-file anchor. | 2 / 3 |
Total | 10 / 12 Passed |