The Meeting-to-Deliverable Pipeline: Recording In, Documents Out
The meeting ends, the recording saves, and the artifact count stays at zero. This pipeline is the fix — a five-skill chain that treats the recording as raw material and ends with documents people actually asked for.
The chain crosses video, writing, and documents categories, because a meeting does too. Order matters; each stage feeds the next.
Stage 1 — understand the call: calesthio-video-understand
Video, 9.2/10. Content-level comprehension of recorded video — what was decided, not just what was said.
Why not straight transcription first: Decision extraction and transcription are different jobs. This one identifies the decision-dense segments, which tells the downstream stages what matters.
Stage 2 — the transcript: baoyu-youtube-transcript
Writing, 9.3/10. Structured transcription — speaker turns, readable paragraphs.
Why this one for meetings: Its formatting is the interface. Speaker-labeled sections are what make the recap skill's output quotable ("as Priya said at 14:20"), and quotability is what makes recaps get read.
Stage 3 — the recap: video-recap
Video, 9.1/10. The summary document: decisions, owners, deadlines, open questions.
The one thing: It's the artifact that changes behavior — if the recap doesn't land in the team channel the same day, the pipeline never runs again. Prioritize its distribution, not its polish.
Stage 4 — the deliverables: create-prd + baoyu-slide-deck
- create-prd (8.1, injection-tested 10/10) — when the meeting produced "we should build this," the PRD skill turns the transcript segment into scoped intent.
- baoyu-slide-deck (9.3, injection-tested 10/10) — when it produced "we need to brief leadership," the deck skill builds it. Passed all four adversarial probes, including declining a destructive-cleanup escalation.
Both held every boundary in our September injection batch — the safe pair at the end of the pipeline.
Stage 5 — the digest: json-canvas
Productivity, 9.3/10. Turns the meeting's output into a canvas/mind-map artifact for the team wiki.
Why it's last: Canvases are navigation, not content. They're only useful once the documents they point at exist.
One recording in, five artifacts out — and the two document generators at the end are the injection-tested ones that held. If your source material is published video rather than internal calls, the content repurposing pipeline is the same idea pointed outward. Video category.