Content
87%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, fully executable, and well-organized for a simple single-purpose skill. Its one material weakness is the absence of validation/error-recovery checkpoints in a workflow that runs a CLI over untrusted document data.
Suggestions
Add a verification step after conversion: check the anydoc/timeout exit status and confirm the output file is non-empty before reporting success.
Add a feedback loop for the documented failure modes (unsupported/image-only input, encryption, malformed content) so Claude detects the specific failure and reports the matching limitation rather than proceeding.
Show how to handle a non-zero exit (e.g., fall back to running anydoc without the timeout wrapper, or surface the failure category to the user) to close the destructive/batch validation gap.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | Lean and efficient with no padding; it never explains what PDF/Markdown is and every line is actionable (quoting rules, commands, security bullets, limits), assuming Claude's competence. | 5 / 5 |
Actionability | Provides fully executable, copy-paste-ready bash (mkdir, mktemp, timeout anydoc, printf), an explicit stdin CSV case, and a concrete hostile-filename quoting example covering the common cases. | 5 / 5 |
Workflow Clarity | A clear sequence is present (mkdir → mktemp → convert → report), but this CLI-on-untrusted-data skill has no validation/verification step (no exit-status or output-file check, no detect-and-recover loop for the listed failure modes), which caps the score at 3 per the destructive-operation guideline. | 3 / 5 |
Progressive Disclosure | Under 50 lines with no external references needed and no bundle files present; the body is well-organized into clear sections, qualifying for the top score under the simple-skill exception. | 5 / 5 |
Total | 18 / 20 Passed |