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.
A strong, example-driven reference: almost every section is executable code or a concrete command sequence, and it stays lean by mostly avoiding background explanation. The main weaknesses are the orphaned references/languages.md file that the body never points to (while duplicating its content), the absence of any validation or error-recovery step in the batch-processing workflow, and one dubious integration example (LangChain's WhisperTranscriptionLoader).
Suggestions
Replace the inline top-10 language list with a signaled one-level reference, e.g. '**Full language list & per-language WER**: See [languages.md](references/languages.md)', so the bundle file is actually reachable and duplication is removed.
Add error handling to the batch-processing example (try/except per file, checking result['text'] is non-empty before writing) so the batch workflow has validation checkpoints and can score above the cap of 3.
Verify or remove the LangChain example — 'WhisperTranscriptionLoader' does not appear to be a real LangChain class; replace it with a working documented pattern or drop the section.
| Dimension | Reasoning | Score |
|---|---|---|
Conciseness | The body is dominated by tight, executable code blocks and tables with minimal prose ('# Requires ffmpeg', model-size table, CLI flag examples), and it largely assumes Claude's competence rather than explaining what Whisper or ASR is. It is not 5 because of padding that earns no tokens — the 'Metrics' block ('72,900+ GitHub stars', 'Trained on 680,000 hours of audio'), a duplicated Resources list, and a top-10 language list that repeats what lives in references/languages.md; not 3 because the waste is minor relative to the whole. | 4 / 5 |
Actionability | Nearly all guidance is copy-paste executable: install commands, load_model/transcribe snippets with result-structure access, per-option examples, full CLI invocations, and a batch loop — matching 'Mostly executable guidance; concrete code or commands with minor gaps'. Not 5 because of gaps: the LangChain example uses 'WhisperTranscriptionLoader' which is not a real documented class, and the batch example has no error handling around individual files. | 4 / 5 |
Workflow Clarity | The single-action core flow (install → load model → transcribe) is unambiguous, but the skill includes batch operations, and the batch-processing section has no validation or error-recovery checkpoint — per the guidelines, missing validation in batch workflows caps workflow clarity at 3. It also stays at 3 rather than 4 because multi-step flows like 'extract audio from video → transcribe → generate subtitles' are scattered across sections with no explicit sequence or checkpoints. | 3 / 5 |
Progressive Disclosure | A bundle file exists (references/languages.md) and the body's Language support section duplicates its content (a top-10 list plus 'Full list: 99 languages total') without ever linking or signaling the reference file — matching 'references present but not clearly signaled; content that should be separate is inline'. Not 4 because an available, well-structured reference is left unnavigable from the SKILL.md; not 2 because the body itself is well-sectioned and the reference is only one level deep. | 3 / 5 |
Total | 14 / 20 Passed |