Webhook contract for transcript → actions

Request (POST)

Minimal body your n8n Webhook node can normalize:

{
  "transcript": "Full meeting text...",
  "meeting_title": "Weekly ops sync",
  "destination": "notion",
  "meeting_date": "2026-09-15"
}

• transcript (required) — plain text; you own transcription upstream
• meeting_title — echoed onto tickets for context
• destination — notion | linear | both
• meeting_date — optional ISO date for relative-date resolution (see null-dates post)

Extracted item shape

{
  "title": string,
  "owner": string | null,
  "due_date": "YYYY-MM-DD" | null,
  "priority": "high" | "medium" | "low" | null,
  "notes": string | null
}

Response sketch

After LLM parse + destination writes, return something humans and machines can both read:

{
  "ok": true,
  "meeting_title": "Weekly ops sync",
  "items": [ /* extracted items */ ],
  "created": {
    "notion_pages": 3,
    "linear_issues": 0
  }
}

On LLM/JSON failure, fail closed: ok: false, no partial tickets if you can help it — better a retry than half a board of junk.

Auth & ops notes

• Protect the webhook (n8n header auth, IP allowlist, or a shared secret query param you rotate).
• Keep LLM, Notion, and Linear credentials in n8n — never in the caller.
• Log meeting_title + item count; avoid logging full transcripts in shared channels if they’re sensitive.

Why a pack instead of a one-off Zap

The hard part isn’t “call an LLM.” It’s a stable schema, destination branching, and prompts that refuse invented dates. Once the contract is fixed, every upstream (Fireflies export, Otter paste, custom bot) can reuse the same webhook.