transcription-skilllisted
Install: claude install-skill zobaerio/zobvia
> **Bundled with Zobvia** (extras/transcription-skill/) — see the sub-skill parts of Zobvia's SKILL.md. Source: [kajisho5/transcription-skill](https://github.com/kajisho5/transcription-skill) v0.13.0 (main @ d8ad0bf), MIT License (LICENSE in this folder). Bundled: SKILL.md, CLAUDE.md, `src/`, `scripts/`, `schemas/`, `pyproject.toml`. Not bundled: docs, tests, evals, brand, assets (hero images). Recognition only — it never invents speech; engine availability is reported by its doctor.
# transcription-skill
Run `transcription <command>` (installed by `pip install ".[faster-whisper]"`), or
`python3 -m transcription_skill.cli <command>`. Every command accepts `--json` (one JSON document on
stdout) and `transcribe` accepts `--dry-run` (plan only, no ASR).
## Workflow
1. **Doctor first** on a new machine: `transcription doctor`. It reports ffmpeg/ffprobe, the engine,
which Whisper models are already cached, the workspace and the cache. Rows are `AVAILABLE`,
`MISSING`, `DEGRADED` or `UNKNOWN`. Do not transcribe when the engine row is MISSING; tell the
user the install line the row shows.
2. **Dry-run before long media**: `transcription transcribe input.mp4 --dry-run --json` shows the
input's duration, the engine and model, the language mode, the budget and whether the cache already
has this exact result (`cache.status: hit` means no ASR will run).
3. **Transcribe**: `transcription transcribe input.mp4 --language ja --json -o input.transcript.json`.
Pass `--lan