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.
A well-orchestrated, instruction-only skill body with an exemplary workflow: explicit checkpoints, state recovery, review feedback loops, and a final checklist. Its weaknesses are token efficiency — duplicated specification blocks and a monolithic 640-line single file — plus a couple of peripheral steps (figures, funded-project search) lacking concrete executable examples.
Suggestions
De-duplicate the body: merge the Constants section with the Parameter Pass-Through table, consolidate the Grant Type Specifications tables with Phase 3's Grant-Specific Drafting Guidelines into one location, and keep the funding/publish track diagram in only one section.
Move the nine per-agency specification tables into per-agency reference files (e.g., references/kakenhi.md, references/nsfc.md) and keep a one-line summary per agency in SKILL.md, enabling progressive disclosure and a leaner overview.
Add concrete executable examples for the vaguer steps: sample WebSearch queries for the KAKEN/NSF/NSFC funded-project lookups, and a minimal matplotlib/SVG snippet for the overview diagram or Gantt chart.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~640-line body is dense with genuinely specialized guidance (agency-specific sections, sub-type age limits, cultural norms) and avoids explaining basics, but contains real duplication: the Constants block duplicates the Parameter Pass-Through table, the Grant Type Specifications tables are restated in Phase 3's Grant-Specific Drafting Guidelines, and the funding/publish track diagram appears twice. Mostly efficient but could be meaningfully tightened, matching anchor 3; not anchor 2 since there is no basic-concept padding or fluffy verbosity. | 3 / 5 |
Actionability | Highly concrete throughout: literal sub-skill invocations ("/research-lit \"$ARGUMENTS\""), full spawn_agent/send_input message templates, the GRANT_STATE.json schema, file-tree output layouts, and matrix/checklist templates. Minor gaps keep it below anchor 5: figure generation ("generate clean SVG or matplotlib directly via code") and the funded-project WebSearch step give no concrete executable examples. | 4 / 5 |
Workflow Clarity | Phases 0-5 are clearly sequenced with explicit 🚦 checkpoints and ⛔ STOP-wait instructions, a state-persistence/resume mechanism for error recovery, a review→revise→re-review feedback loop with severity levels (CRITICAL/MAJOR/MINOR), and a final pre-completion checklist — matching anchor 5 (explicit validation steps, feedback loops, checklists). | 5 / 5 |
Progressive Disclosure | No bundle files exist (no references/, scripts/, or assets/), and the nine per-agency specification tables (~75+ lines of detail) are inlined monolithically where per-agency reference files would keep the SKILL.md body lean. Internal heading structure is good and the shared-references links are clearly signaled, placing it above anchor 2's 'minimal structure' but squarely at anchor 3's 'content that should be separate is inline'. | 3 / 5 |
Total | 15 / 20 Passed |