Content
42%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 has real bones — verified bundle files, executable audit commands, and a concrete CLI parameter table — but roughly three-quarters of it is duplicated generic policy boilerplate, and it never explains how to actually write a budget justification (no narrative examples, no per-category guidance, no agency-specific requirements). The single reference is properly wired but too thin to carry the skill's substance.
Suggestions
Consolidate the six overlapping policy sections (Error Handling, Input Validation, When Not to Use, Required Inputs, Output Contract, Failure Handling) into one short section and delete the generic template blocks (Risk Assessment, Security Checklist, Evaluation Criteria, Lifecycle Status, Response Template) — 'Quick Check' is a duplicate of 'Audit-Ready Commands' and the first 'When to Use' bullet awkwardly re-pastes the description verbatim.
Add the skill's actual substance: a worked example of a narrative justification (e.g. the $50,000 mass spectrometer case written out in full) and per-category guidance for the listed Use Cases (Equipment, Personnel, Supplies, Travel), since currently only the script's internals know how to produce the deliverable.
Move agency-specific compliance details (NIH vs NSF requirements) into references/ and point to them from a single References section, keeping SKILL.md as a lean overview of when to use the skill and how to invoke the script.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is heavily padded with boilerplate: 'Quick Check' duplicates 'Audit-Ready Commands' (both run `python -m py_compile scripts/main.py`), and at least six sections restate the same fallback policy ('Error Handling', 'Input Validation', 'When Not to Use', 'Required Inputs', 'Output Contract', 'Failure Handling', 'User Checkpoints'). Generic template sections ('Risk Assessment', 'Security Checklist', 'Evaluation Criteria', 'Lifecycle Status', 'Response Template') add no skill-specific value, matching 'Noticeably verbose; several unnecessary explanations or padded sections'. It avoids a 1 only because no section explains concepts Claude already knows and individual sections are short. | 2 / 5 |
Actionability | Concrete elements exist — the executable commands (`python scripts/main.py --help`, `py_compile`), a real Parameters table for the CLI, and a verified reference file — but guidance for the skill's actual deliverable is missing: there is no example justification text, no narrative structure to follow, and the Example section is a single vague line ('Input: $50,000 for mass spectrometer / Output: Justification emphasizing essentiality and cost-sharing'). This fits 'Some concrete guidance but incomplete; missing key details'. | 3 / 5 |
Workflow Clarity | The 5-step Workflow is sequenced and includes stop conditions ('stop early if the task would require unsupported assumptions') and a fallback path ('switch to the fallback path and state exactly what blocked full completion'), plus a separate 'Quick Validation' section. However, the steps are generic process boilerplate not tied to this skill's operations — no command invocations embedded in the sequence, no validate-output-against-agency-requirements step, and checkpoints are scattered across redundant sections rather than integrated. 'Steps listed but validation gaps; checkpoints missing or implicit' is the closest anchor. | 3 / 5 |
Progressive Disclosure | Bundle structure is real and correctly linked: references/audit-reference.md exists, is one level deep, and is clearly signaled in a References section, and scripts/main.py exists as the documented entry point. But the SKILL.md inlines ~150 lines of template boilerplate (risk tables, checklists, lifecycle metadata like 'Next Review Date: 2026-03-06') that should be pruned or split, while the genuinely detailed material (how to write each justification type) is absent from both the body and the single thin reference. This matches 'Some structure but could be better organized; content that should be separate is inline'. | 3 / 5 |
Total | 11 / 20 Passed |