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.
The body contains genuinely useful, accurate task content (CLI parameters, data formats, journal styles, quality checklists) that matches the real script API, but it is wrapped in ~150 lines of generic template boilerplate and duplicated sections. It is a monolithic 530-line document that barely uses its own references directory.
Suggestions
Cut the generic template sections (Key Features, Implementation Details, Output Requirements, Response Template, Inputs to Collect, Output Contract, Validation and Safety Rules, Lifecycle Status) — they are skill-agnostic filler; consolidate the two duplicate dependency listings.
Make CLI examples copy-paste executable by removing the "# Example invocation:" comment prefix, and drop the hardcoded `cd "20260318/scientific-skills/..."` path.
Link references/runtime_checklist.md by name, and move the journal style configuration JSON, journal-specific notes, and FAQ into reference files to shrink SKILL.md to an overview.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body runs ~530 lines with substantial template boilerplate ("Key Features", "Implementation Details", "Output Requirements", "Response Template", "Inputs to Collect", "Output Contract", "Validation and Safety Rules") that teaches Claude nothing skill-specific. There are also redundancies — two dependency listings (a "Dependencies" section listing both `pil` and `pillow`, plus a separate "Dependency Requirements" block) and broken self-references ("See `## Prerequisites` above" / "See `## Usage` above" pointing to sections that appear later or duplicate content). Genuine task content (parameter tables, style configs) is buried in the padding. | 2 / 5 |
Actionability | Concrete, verified guidance dominates: `python -m py_compile scripts/main.py` is executable; CLI examples use flags that match the actual argparse surface in scripts/main.py (--input, --time-col, --event-col, --group-col, --style, --time-points, --combine, --km-plot); and the Python API example matches the real `RiskTableGenerator` class (main.py:41). The gap keeping it below 5: every CLI example is commented out ("# Example invocation:" prefix) rather than copy-paste runnable, and the Example Usage block begins with a bogus hardcoded path (`cd "20260318/scientific-skills/..."`). | 4 / 5 |
Workflow Clarity | A clear sequence exists with explicit checkpoints: the Workflow section's 5 steps include validation ("stop early if the task would require unsupported assumptions") and a fallback path ("If execution fails ... switch to the fallback path and state exactly what blocked full completion"); "Quick Check" supplies a concrete pre-execution validation command; and "Error Handling" defines failure reporting per failure mode. Minor gaps: three overlapping workflow descriptions ("Example run plan", "Workflow", and an unlinked references/runtime_checklist.md) dilute which path to follow, and validation steps are described generically rather than woven into the CLI usage sequence. | 4 / 5 |
Progressive Disclosure | Section headers exist and the body points to real bundle files (scripts/main.py, references/), but it is a ~530-line monolith: journal style configs, the algorithm description, FAQ, journal-specific notes, lifecycle status, and security checklists are all inlined content that belongs in reference files. The single reference file (references/runtime_checklist.md) is never named or linked — the body says only "Reference material available in `references/`" — so references are present but not clearly signaled. | 3 / 5 |
Total | 13 / 20 Passed |