scribe
/scribe
Turn meeting recordings into notes with an unbroken chain of custody from audio to claim. Use when the user has recordings or audio to transcribe or summarize, or when an existing transcript or auto-summary reads as garbled or untrustworthy.
- Author
- ooiyeefei ↗
- Repository
- ccc ↗
- Licence
- MIT
- Last repository change
- 15.10.2026
Assessment and security
- Files in this package
- 7
- Scripts
- 4
- HTML comments
- 0
- URLs in SKILL.md
- 0
No signals detected. Repository changed on 15.10.2026, without references to retired models or APIs.
MIT, open: use and modify while retaining author attribution.
- 4 scripts in this package. Review them before installing, as Claude may execute them.
Scripts: scripts/_thresholds.py, scripts/audit.py, scripts/transcribe.py, tests/test_provider_selection.py
What it does and when to use it
Turn meeting recordings into notes with an unbroken chain of custody from audio to claim. Use when the user has recordings or audio to transcribe or summarize, or when an existing transcript or auto-summary reads as garbled or untrustworthy.
Original author description
Turn meeting recordings into notes with an unbroken chain of custody from audio to claim. Use when the user has recordings or audio to transcribe or summarize, or when an existing transcript or auto-summary reads as garbled or untrustworthy.
How it works
- 1. Establish the recording's context
- 2. Transcribe
- 3. Audit before you read
- 4. Corroborate against any second transcript
- 5. Write the notes
SKILL.md preview
--- name: scribe description: Turn meeting recordings into notes with an unbroken chain of custody from audio to claim. Use when the user has recordings or audio to transcribe or summarize, or when an existing transcript or auto-summary reads as garbled or untrustworthy. --- A summariser cannot hear. Hand it a garbled span and it **launders** the noise into a clean fact — "some paying customer" becomes "~8,000 users" — and nothing downstream can separate that from a real figure. This skill holds a **chain of custody**: every claim in the notes traces back to audio the model actually heard well, and a claim whose custody is broken says so. Scripts are in `scripts/`. They need a Python with `httpx`; `--provider local` also needs `faster-whisper`. ## 1. Establish the recording's context Ask the user, or read it off the surrounding material — the files, the repo, an existing transcript: - **Language** — the ISO code to pin. Left unpinned, per-window detection flaps on code-switched speech and the decoder emits fluent, confident text in the wrong language. - **Speakers** — how many, plus names and roles. - **Domain vocabulary** — product names, people, companies, jargon, currencies. Done when you can state the language code, the speaker count, and at least five domain terms. ## 2. Transcribe ``` python scripts/transcribe.py AUDIO... --out DIR --language <code> --keyterms "term,term,..." ``` `--provider local` is the default; `auto` is a compatibility alias for local. Existing API keys do not trigger an upload. Select `elevenlabs`, `deepgram` or `openai` explicitly when hosted transcription is appropriate. Setup, capability and cost per provider: [`references/providers.md`](references/providers.md). Start with local faster-whisper. It does not diarize, so speaker attribution stays uncertain unless corroborated. When speaker labels are required, choose an authorised hosted diarizing provider explicitly and add `--diarize`. A missing local dependency fails without switching to a hosted provider. Pre-cache weights for fully offline use. Done when every input file has a `.json` and `.txt` in DIR. ## 3. Audit before you read ``` python scripts/audit.py DIR ``` Two findings, both of which vanish once a transcript is flattened into prose: - **Gaps** — a recorder stopped mid-meeting drops content in silence, and the notes that follow read as complete. A gap is a stretch of the meeting you hold no evidence for. - **Risky spans** — numbers and proper nouns resting on low-confidence audio. This is where laundering happens. Give every finding one of three dispositions: corroborated against a second transcript, confirmed with the user, or carried into the notes as a marked uncertainty. Done when the count of dispositions equals the count of findings. ## 4. Corroborate against any second transcript When the user has another transcript of the same audio — Granola, Otter, an earlier run — align it against yours. Two independent decodes are the cheapest confidence signal available. A claim present in one and absent from the other is a divergence, not a fact, and inherits the lower confidence of the two. Where your audio has gaps, the other transcript may be the only evidence that exists; anything sourced that way stays marked as uncorroborated. ## 5. Write the notes Group by topic, so each section stands on its own. Every figure and every named entity carries its custody: | Traced to | Written as | |---|---| | a span above threshold | stated plainly | | a flagged span | stated with its marking and timestamp | | a gap-filling second transcript only | marked uncorroborated | | no source | absent | Cite timestamps for anything a reader might challenge. Done when every figure and named entity in the notes falls into one of those four rows.
Files in this package (7)
- README.md 7 KB
- SKILL.md 4 KB
- references/providers.md 6 KB
- scripts/_thresholds.py 2 KB
- scripts/audit.py 10 KB
- scripts/transcribe.py 16 KB
- tests/test_provider_selection.py 4 KB
Versions
| Repository | Score | Licence |
|---|---|---|
| This page · ooiyeefei/ccc | 100 | MIT |
| simota/agent-skills | 100 | MIT |
| queelius/claude-anvil | 100 | MIT |
| zoom/skills | 100 | MIT |
| danielgouveiah/glasses | 100 | MIT |
| Trevojan/regency | 86 | MIT |
Database snapshot: 13.11.2026 · package version 5a949076f8eb (SHA-256 digest of the entire package)
Downloads in the last 30 days: 0 · ranking
Report a problem with this skill: harmful content, incorrect descriptions or an author’s removal request.
Similar skills
- 100speech-to-text
Transcribe audio to text using ElevenLabs Scribe v2. Use when converting audio/video to text, generating subtitles, transcribing meetings, or processing spoken content. - 100local-transcribe
Transcribe local audio/video files to Markdown via ffmpeg + ElevenLabs Scribe. Use whenever the user provides paths to local audio/video files (mp3, mp4, m4a, wav, mov, mkv, etc.) and wants raw transcription without summarization. Triggers on: paths with audio/video extensions +… - 90transcribe-recording
Transcribe audio and video of lectures, meetings, talks, interviews, and other conversations with timestamps, check speaker separation, and prepare context inputs. Use for speech recordings; route existing text without audio to build-agent-context. - 100speech-to-text
Transcribe audio to text using ElevenLabs Scribe v2. Use when converting audio/video to text, generating subtitles, transcribing meetings, or processing spoken content. - 100meeting-summarizer
Transforms raw meeting notes, transcripts, or recordings text into a concise, scannable summary with clear sections for decisions made and owned action items (owner + due date). Use this skill when the user asks to "summarize this meeting", "turn these notes into a summary",… - 90meeting-notes
Turns raw, messy meeting notes or a transcript into a crisp, shareable summary. Use when someone says "summarize these meeting notes", "turn this transcript into action items", "what did we decide", "clean up my notes", or "send a recap". Works for any meeting type — standup,…