Content
60%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 delivers a well-structured, actionable orchestration: explicit step sequence, on-demand loading of six real reference files, resume capability, and error-retry loops. Its weaknesses are heavy redundancy (the lazy-loading rule and the Implementation Requirements section repeat earlier content almost verbatim) and internal inconsistencies in step numbering and one stale filename reference.
Suggestions
Delete the "Implementation Requirements" section and consolidate the on-demand-loading rule into one statement; it is repeated roughly six times across the document.
Fix the stale filename: the Architecture section lists "05-generate-document.md # NIH Protocol generation" but the actual file and step mapping use "references/05-concatenate-protocol.md".
Unify step numbering: the Full Workflow Logic restarts at "Step 1/2/3" (with a "Step 2.5" menu) inside the global Steps 0-5 sequence, making references like "Continue to Step 3 (Final Summary)" ambiguous.
Add a minimal usage example for scripts/sample_size_calculator.py (or a pointer to the subskill that documents it) so the sole script is actionable from SKILL.md.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The on-demand loading rule is stated at least five separate times ("Do NOT pre-read subskill files", "ON-DEMAND LOADING", "Do NOT read subskill files in advance or 'to prepare'", "LAZY LOADING - On-Demand Only"), and the entire "Implementation Requirements" section restates the "Execution Flow" section nearly point-for-point. This is noticeably verbose with several padded, redundant sections, matching the level-2 anchor rather than the level-3 'mostly efficient' anchor. | 2 / 5 |
Actionability | Concrete, executable guidance throughout: the exact welcome message to display, exact menus with option-by-option logic, an explicit step-to-file mapping ("Step 2: references/02-protocol-foundation.md ... sections 1-6"), install commands ("pip install -r requirements.txt"), a concrete waypoint file layout, and a research-summary template. It stops short of 5 because the script is never shown in use ("scripts/sample_size_calculator.py # Statistical power analysis (validated)") and placeholders like "[TBD] items" are left unexplained. | 4 / 5 |
Workflow Clarity | Steps 0-5 are clearly sequenced with completion-status checks, resume-from-waypoint logic, a retry/exit feedback loop on step failure, and an MCP connectivity check in Step 1. It is not a 5 because of numbering incoherence: the Full Workflow Logic restarts its own Step 1/2/3 numbering inside the global 0-5 sequence ("Step 2.5: Protocol Completion Menu", "Continue to Step 3 (Final Summary)"), and the architecture section lists "05-generate-document.md # NIH Protocol generation" while the mapping and the actual filesystem say "references/05-concatenate-protocol.md". | 4 / 5 |
Progressive Disclosure | The bundle structure is genuinely good: six one-level-deep subskill files in references/ (verified to exist on disk), each clearly signaled by step-to-file mapping and loaded on demand, plus a separate scripts/ calculator and assets/ template. It is not a 5 because the ~500-line body inlines full menus, message templates, and a research-summary markdown template that duplicate detail more appropriate to the referenced subskills, keeping SKILL.md from being a lean overview. | 4 / 5 |
Total | 14 / 20 Passed |