Content
56%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 strong executable core — runnable commands, a complete parameter table, design-type/spending-function references, and a concrete output example — buried in heavy audit boilerplate. Roughly 100+ lines of meta-sections, four overlapping error-handling sections, and a References list naming nonexistent files hurt token efficiency and navigation.
Suggestions
Cut the meta-boilerplate sections (Risk Assessment, Security Checklist, Lifecycle Status, Evaluation Criteria, Technical Difficulty) and merge the four overlapping scope/error sections (Error Handling, Failure Handling, Input Validation, When Not to Use) into a single section — this alone removes roughly 40% of the body's tokens.
Delete the second 'References' section that lists files not present in references/ (keep only the working audit-reference.md link), and move the Parameters, Design Types, and Spending Functions tables into a bundled reference file so SKILL.md stays a lean overview.
Bundle an actual requirements.txt, or replace "pip install -r requirements.txt" with "pip install numpy scipy matplotlib", so every documented command runs against the real bundle.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The ~305-line body is padded with meta-boilerplate sections that teach Claude nothing ("Risk Assessment", "Security Checklist", "Lifecycle Status" with a next-review date, "Technical Difficulty: **HIGH**", "Evaluation Criteria"), four overlapping error/scope sections ("Error Handling", "Failure Handling", "Input Validation", "When Not to Use"), and commands repeated up to three times ("python -m py_compile scripts/main.py" appears in both Quick Check and Audit-Ready Commands). That is several unnecessary padded sections (score 2), though not 1 since it never explains concepts Claude already knows and the core usage content is dense. | 2 / 5 |
Actionability | Concrete, copy-paste-ready commands ("python scripts/main.py --design adaptive_reestimate --n-simulations 25 --optimize") plus a complete parameter table with defaults, design-type and spending-function tables, and a realistic JSON output example. Not 5 because of minor gaps: "pip install -r requirements.txt" targets a requirements.txt that is not present in the bundle. | 4 / 5 |
Workflow Clarity | The Workflow section gives a sequenced 5-step path with an explicit feedback/fallback loop ("If execution fails or inputs are incomplete, switch to the fallback path and state exactly what blocked full completion"), backed by explicit checkpoints (Quick Check py_compile, "python scripts/main.py --help", the Quick Validation checklist). Not 5 because the checkpoints are scattered across redundant sections and the steps are governance-level rather than a concrete simulate-then-interpret-then-optimize procedure. | 4 / 5 |
Progressive Disclosure | The one real reference is well-signaled ("[references/audit-reference.md](references/audit-reference.md) - Audit-ready assumptions, supported design modes, and fallback boundaries") and scripts/main.py exists as bundled, but a second "References" section lists phantom entries ("Adaptive design statistical theory", "Regulatory guidance documents", "Alpha spending function literature") that match no file in references/, and audit/lifecycle content that belongs in a separate reference is inlined. Some structure, but organization gaps and misleading navigation keep this at 3 rather than 4. | 3 / 5 |
Total | 13 / 20 Passed |