Content
50%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 skill has a solid executable core — real script, concrete commands, parameter table, strategy guide — but it is buried under template-generated boilerplate, self-referential filler, duplicated content, and references that don't deliver what the body promises. A significant cleanup pass would cut the file roughly in half and sharpen the workflow.
Suggestions
Remove the template filler: the verbatim description repeats in "When to Use"/"Key Features", the "See `## X` above" pointer sections, the empty "Dependencies" header, the contradictory pip install line, and the generic "Output Requirements"/"Response Template"/"Input Validation" sections.
Wire validation into the trimming workflow: after trimming, run a check (word count plus `--check-only` on output, and a key-finding/P-value preservation check), and add per-file error handling to the batch loop so the batch cap no longer applies.
Make `references/guidelines.md` deliver the content the body promises (compression strategies, protected elements like numbers/P-values/CIs, journal word limits), or stop advertising those documents, and drop the hardcoded `20260318/scientific-skills/...` path from Example Usage in favor of a relative path from the skill root.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is noticeably verbose and padded: the frontmatter description is pasted verbatim into "When to Use" and "Key Features"; sections point at themselves ("See `## Features` above for related details"); "Dependencies" is an empty header; `pip install -r requirements.txt` is immediately contradicted by "No external dependencies required"; the py_compile command appears three times; and large generic boilerplate sections ("Output Requirements", "Response Template", "Input Validation", "Error Handling") add tokens unrelated to abstract trimming. Not 1 because genuine concrete content (parameter table, executable commands, strategy table) is present; not 3 because the padding and template filler are pervasive, not isolated. | 2 / 5 |
Actionability | Concrete, executable commands are provided against a real packaged script (e.g. `python scripts/main.py --text "..." --target 40 --check-only`), plus a complete parameter table, output format examples, and a batch loop. Not 5 because the "Example Usage" block opens with a hardcoded non-portable path (`cd "20260318/scientific-skills/..."`), and some guidance remains abstract ("Return a structured result that separates assumptions, deliverables, risks..."). | 4 / 5 |
Workflow Clarity | A sequence is present (Workflow section, Example run plan) with some checkpoints (py_compile Quick Check, `--check-only` mode, an error-handling fallback), but the verify-after-trim step is only a listed "Quality Validation" feature and is never wired into the workflow, and the batch example has no per-file validation or error handling — the batch-operation cap applies. Not 4 because the core trim-then-validate feedback loop is missing or implicit. | 3 / 5 |
Progressive Disclosure | The bundle structure is real and signaled (references/ and scripts/main.py exist and are referenced), but `references/guidelines.md` is a 7-line stub that does not contain the three documents the body promises ("Compression strategies documentation", "Protected elements guidelines", "Journal word limits by publisher"), and much generic boilerplate that belongs in no file at all is inlined in SKILL.md. Not 4 because the promised reference content does not match the actual bundle and the main file carries content that should be split or removed. | 3 / 5 |
Total | 12 / 20 Passed |