ingest-meeting

name	ingest-meeting
description	Classify a meeting transcript, create brain/artifact outputs with source links, extract knowledge, update the ingest log and changelog.

/ingest-meeting

Purpose

Process a meeting transcript from inbox/meetings/ into durable knowledge: meeting summary in logs/meetings/, updates to person and project pages, action item tasks, memory events for durable decisions, and hub updates for named entities.

When to use

• A new transcript landed in inbox/meetings/.
• A self-reflection / interview transcript landed (handled specially below).

Arguments

• <transcript-path> (required) — vault-relative path. e.g., inbox/meetings/2026-05-28-q3-planning.md.

Steps

1. Read and classify.

   • Regular meeting — discussion about work. Two or more participants.
   • Self-reflection / interview — the user thinking aloud, structured against a prompt. Handle specially (see below).
   • Not relevant — small talk, accidental capture. Archive and exit.

2. Speaker resolution. Transcripts often have generic "Speaker 1 / 2 / 3" labels. Resolve to actual attendees by:

   • Cross-referencing the calendar event for that date/time.
   • Looking at the email header (if extracted from a call platform).
   • Asking the user if unclear.

3. For a regular meeting:

   a. Create the meeting summary at logs/meetings/YYYY-MM-DD-<slug>.html:

   • Robin v0.2 HTML.
   • <meta name="robin:type" content="meeting">.
   • <meta name="robin:date" content="YYYY-MM-DD">.
   • <meta name="robin:attendee" content="<name>"> repeated for each attendee.
   • <meta name="robin:duration" content="N min">.
   • <meta name="robin:source" content="inbox/meetings/...">.
   • Body: one-line summary, key points (3–8 bullets), decisions made, action items, follow-ups, link to transcript.

   b. Extract knowledge:

   • Update each attendee's page in brain/people/<bucket>/ with anything new about them.
   • Update the relevant project page in brain/projects/<slug>/ with status changes, decisions, scope shifts.
   • Create or update decisions in brain/decisions/YYYY-MM-DD-<slug>.html when explicit decisions were made.
   • Create or update patterns/standards/playbooks when generalizable rules surfaced.
   • For each, add provenance: <meta name="robin:source" content="inbox/meetings/..."> and inline citation per ../lib/provenance.md.

   c. Memory events for compact durable cues (preferences, corrections, source-of-truth notes) via memory.save.

   d. Tasks. For each action item, invoke /create-task with:

   • Owner inferred from context (the attendee who committed).
   • Source: meeting:logs/meetings/YYYY-MM-DD-<slug>.html.
   • Project: the project discussed.

   e. Cross-link. The meeting page links to: each attendee's page, the project page(s) discussed, each decision created, each task created. Each linked entity in turn references the meeting page in its own backlinks (computed at read time).

   f. Hub update (mandatory). Any new tool / vendor / framework / org mentioned must go on a hub.

   g. Archive the transcript. Move inbox/meetings/<file> → inbox/archived/meetings/<file>.

4. For a self-reflection / interview:

   Self-reflections are sensitive. Different rules:

   a. No meeting page. Self-reflections don't become public-style summaries.

   b. Memory extraction preferred. Extract preferences, decisions, working-style insights as memory events. Tier them semantic or procedural as appropriate.

   c. Update brain/about_user/ with insights about communication style, values, working hours, blindspots. Be careful — these pages are personal context.

   d. Update project pages if the reflection surfaced project insights (cautiously — preserve the reflective tone).

   e. Compact, not broad. When a reflection contains sensitive content, prefer compact actionable memory over broad prose. Avoid recording emotional content as durable text unless the user explicitly asks.

   f. Archive the transcript as with regular meetings.

5. Update tracking:

   • Append to logs/ingest-log.md:

     ## YYYY-MM-DDTHH:MM:SSZ — meeting — <slug>
     source: <inbox-path>
     outputs: logs/meetings/..., brain/decisions/..., brain/tasks/...
     attendees: [name1, name2]
     

   • Append to logs/changelog.md:

     ## [YYYY-MM-DD] ingest | meeting [[<slug>]] — <one-line summary>
     

6. Report:

   Type: meeting / self-reflection
   Created: <meeting page or memory events>
   Updated: <project pages, person pages>
   Tasks: N
   Memory events: N
   Hubs touched: <list>
   Archived: inbox/archived/meetings/<file>
   

Output shape

Concise. Show the user what was extracted and where it went. They should be able to spot-check by clicking through.

Edge cases

• Speaker labels can't be resolved. Ask the user. Don't invent attribution.
• Mixed content (e.g., a meeting that ended in self-reflection). Process the regular-meeting part normally; flag the reflection portion for follow-up.
• Sensitive content (HR, personal). Be conservative. Compact actionable memory only, with sensitivity: private on any pages or memory events. Surface to user before publishing widely-linked content.
• The transcript is just a recording, not transcribed. Halt and tell the user. Don't try to transcribe.
• A decision is implicit (not stated outright). Surface as a possible decision in the report; let the user confirm before creating a decision page.