Content
65%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 is lean and dense, with copy-paste-ready pdf2md examples, strong consent/credential gating, and wise deferral of MinerU basics to the official skill. The main weaknesses are the absence of a sequenced workflow with post-conversion validation for batch runs, and pointers (`reference.md`, `setup.sh`) that do not resolve in the actual bundle.
Suggestions
Add an explicit ordered workflow (install → `scripts/check.sh` → `pdf2md`) with a validation checkpoint after batch/directory conversion, e.g. verify one output .md exists per input document before reporting success, to lift workflow_clarity past the batch-operations cap.
Ship `reference.md` (and `setup.sh`) in the bundle or fix the pointer — currently "Details, endpoints and troubleshooting are in `reference.md`" references a file that does not exist next to SKILL.md, breaking navigation.
Include one concrete MCP example (e.g. a `parse_documents` call and its `OUTPUT_DIR` result) to close the actionability gap for the cloud path.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean and assumes competence: "For reading, searching, locators, tiers and CLI usage of MinerU itself, use the official `mineru` skill... Do not restate that material here." Not 5 because of small trimmable asides ("A `/api/v4` URL in the cloud API is unrelated to the local package version", "There are two different MinerU variables") and inline version pins; not 3 because there is no padding and no explanation of concepts Claude already knows. | 4 / 5 |
Actionability | Copy-paste-ready commands cover the common cases: five `pdf2md` invocations (PDF, scanned+OCR, page range, directory, URL), `with-secret [--optional] NAME -- cmd`, and `bash scripts/check.sh [--online]`. Not 5 because there is no example MCP `parse_documents` call and no concrete `setup.sh` invocation; not 3 because the guidance is overwhelmingly executable rather than pseudocode or high-level hints. | 4 / 5 |
Workflow Clarity | Sections are topical rather than a sequenced workflow ("## Install" appears last), and the batch operation "pdf2md ./papers ./papers_md" has no output validation or fix-and-retry loop, so the batch-operations cap applies. Not 4 because although `check.sh` is a genuine validation checkpoint ("reports the install state, model readiness...") and consent gating is explicit, per-run verification is missing and the overall sequence is implicit; not 2 because each section does define its steps concretely. | 3 / 5 |
Progressive Disclosure | The ~45-line body is well sectioned and correctly defers detail ("Details, endpoints and troubleshooting are in `reference.md`"), but scoring against the actual bundle, `reference.md` and `setup.sh` do not exist — only `scripts/check.sh` is present, so the primary deep-dive pointer does not resolve. Not 4 because a clearly signaled reference that 404s is more than a minor organization gap; not 2 because the body is well organized and self-sufficient for core usage with one-level-deep pointers. | 3 / 5 |
Total | 14 / 20 Passed |