From devboy
Fetches speaker-attributed, timestamped meeting transcripts—full, paginated, or filtered by speaker/phrase—without leaking raw PII into storage. Use for specific quotes or targeted pulls from large transcripts.
npx claudepluginhub meteora-pro/devboy-tools --plugin devboyThis skill uses the workspace's default tool permissions.
Retrieve the speaker-attributed, timestamped transcript for a single meeting. Transcripts are routinely large (thousands of sentences for a one-hour call), so the skill is built around pulling only what the question actually needs rather than the whole thing.
Queries Fireflies.ai for meeting transcripts from Zoom, Teams, Google Meet; searches by keyword/date, extracts action items, and retrieves summaries.
Fetches Fireflies.ai meeting transcripts via GraphQL, processes speaker-diarized sentences, extracts action items and summaries for intelligence pipelines.
Downloads Zoom meeting transcripts to VTT files via CLI using meeting ID. Outputs timestamped text with speaker labels. Requires Zoom OAuth env vars and hamel package.
Share bugs, ideas, or general feedback.
Retrieve the speaker-attributed, timestamped transcript for a single meeting. Transcripts are routinely large (thousands of sentences for a one-hour call), so the skill is built around pulling only what the question actually needs rather than the whole thing.
meeting-to-tasks) needs the raw text to extract action items that are not already in action_items.If the user does not yet know which meeting — route through meeting-search first.
You need a meeting_id. If the user already pasted one, skip this step. Otherwise confirm the id via get_meeting_notes (or via meeting-search for a keyword lookup) and pin the id before fetching:
devboy tools call get_meeting_notes '{
"from_date": "2026-04-16T00:00:00Z",
"to_date": "2026-04-17T00:00:00Z",
"limit": 10
}'
Confirm the title and date with the user before proceeding — pulling the wrong transcript is an irreversible PII leak into the chat.
devboy tools call get_meeting_transcript '{"meeting_id": "<id>"}'
devboy tools call takes a tool name + a JSON argument object — it does not accept --budget (or any other CLI flag) for the tool itself. The format-pipeline budget is applied inside the tool runtime. The transcript output is rendered as plain text, not JSON; a long transcript may be shortened by the pipeline and picked up again via a subsequent call.
If the response ends mid-thought, pull the next segment:
devboy tools call get_meeting_transcript '{"meeting_id": "<id>", "chunk": 2}'
chunk is a per-call argument understood by the transcript tool (the default is 1). The response is still text — do not expect a JSON body with fields like total_chunks or chunk_number. The tell for "more to fetch" is that the text ends mid-sentence, not a structured field.
When the question is specific ("what did Alice say about X?"), do not scroll the whole transcript in the chat. Two patterns:
Search first, transcript second. Run meeting-search with the topic word, confirm this is the right meeting, then open the relevant chunk.
Grep the chunk. The transcript is rendered as text lines shaped like [mm:ss] <speaker>: <sentence>, so filter with grep / ripgrep:
devboy tools call get_meeting_transcript '{"meeting_id": "<id>"}' \
| grep -F '] Alice: '
When you respond to the user, reply with the minimum quote that answers the question — typically one to three sentences plus the speaker and timestamp. Never paste the full transcript into the chat unless the user explicitly asked for it.
meeting_id leaks a different meeting's contents into the chat and is not retractable.summary and action_items fields returned by get_meeting_notes — those are provider-generated and cheaper to fetch than the full transcript.