Pka meetings

pka-meetings

Full meeting documentation pipeline for a Personal Knowledge Assistance system.

Integration Boundary

pka-meetings orchestrates thinkkit skills when available, but the boundary is clear: thinkkit handles capture and reconciliation; PKA handles routing, indexing, attendee linking, and persistence. Neither repo modifies the other. thinkkit remains fully functional without PKA installed.

thinkkit                              PKA
────────                              ───
take-notes ──────────────────────▶  pka-meetings ──▶ knowledge/
                                          │
resolve-against-transcript ──────▶        ├──▶ .pka/session-log.md
                                          └──▶ .pka/knowledge.db

Prerequisites

Both thinkkit and pka-skills should be installed. pka-meetings checks for thinkkit skills at startup and degrades gracefully if absent:

• No take-notes: the skill still handles routing, indexing, and logging of manually-written notes
• No resolve-against-transcript: the skill skips reconciliation and routes notes as-is
• Both absent: the skill functions as a PKA-aware meeting note router only

Pre-Flight

1. Load .pkaignore and Repo Map from CLAUDE.md
2. Check for take-notes: ls .claude/skills/take-notes/ 2>/dev/null
3. Check for resolve-against-transcript: ls .claude/skills/resolve-against-transcript/ 2>/dev/null
4. Check for .pka/knowledge.db
5. Scan team-inbox/ for transcript files (patterns in references/transcript-patterns.md)
6. Scan for earshot capture directories — any directory containing a meeting.json marker file (typically under meeting-notes/, but the check is marker-based, not path-based; see references/transcript-patterns.md). A capture containing transcript.md with no reconciled note yet is a Mode 2 candidate, same as a detected Zoom transcript.

If thinkkit skills are absent, report which modes are unavailable but proceed with route-only mode.

Mode 1: Capture (Live Note-Taking)

Trigger: User says they're in a meeting, or says "take notes", or invokes /thinkkit:take-notes with PKA context active.

1. Ask: "What's this meeting?" and "Who's attending?"
2. Invoke take-notes — user feeds terse observations; skill expands into structured notes
3. Retrieve output notes file
4. Run Post-Processing Pipeline

Mode 2: Reconcile Against Transcript

Trigger: User mentions a transcript, a transcript file is detected in team-inbox/, or an earshot capture directory with an unreconciled transcript.md is detected.

1. Identify transcript (from team-inbox/ detection, an earshot capture's transcript.md, or user-specified path)
2. Find matching notes (today's notes, or ask user to specify; if none: offer to generate notes from transcript directly). For earshot captures, prefer the notes file whose **Recording**: header points at the capture directory.
3. Invoke resolve-against-transcript with notes + transcript
4. Run Post-Processing Pipeline
5. Ask about transcript disposition: keep in team-inbox/ / archive to knowledge destination / delete (confirm before delete). Applies to loose team-inbox/ transcripts only — earshot capture artifacts stay inside their capture directory; never offer to move or delete them.

Ordering: reconciliation (and narration, if the user runs earshot's /meeting-narrative) happens before routing, while the notes are still in meeting-notes/ — meeting-narrative finds notes by globbing there. If the user wants a narrative, suggest generating it now, before Step 4 moves the notes.

Mode 3: Route Only

Trigger: User provides an existing notes file or says "file these meeting notes".

Skip directly to Post-Processing Pipeline with the provided notes file.

Post-Processing Pipeline

Runs after any capture or reconciliation, or directly for route-only mode.

Step 1 — Infer meeting type and routing destination

Using meeting title, attendees, and Repo Map. See references/meeting-routing-rules.md for full logic.

Priority order:

1. 1-1 meeting (title contains "1-1"/"one-on-one", or single attendee other than user) → <personnel-folder>/<person>/ if that folder exists per Repo Map
2. Active project meeting (title or attendees match a known project slug) → ask: "Route to projects/<n>/ now, or to knowledge/ after the project wraps up?"
3. Leadership/strategy (strategic keywords in title, or executive attendees) → meeting-home-tagged leadership folder per Repo Map
4. General → any meeting-home-tagged folder; if none, ask

Present proposed destination. Confirm before moving (except hands-off autonomy).

Step 2 — Attendee linking

See references/attendee-linking.md for matching logic.

For each attendee extracted from notes:

• Fuzzy match against Repo Map personnel folder names (first name, last name, slug)
• If match: add relative path reference under ## Attendees at bottom of notes
• If no match: list name without link

Step 3 — Action item extraction

See references/action-item-patterns.md for detection patterns.

Scan notes for:

• Lines starting with - [ ], →, AI:, Action:, TODO:
• Sections ## Action Items, ## Next Steps, ## Follow-up

Normalize into ## Action Items section at bottom:

- [ ] <action> — @<owner if mentioned> — <due date if mentioned>

Surface to user: "Found N action items. Add to session log open threads?"

Step 4 — Save to destination

File naming: <YYYY-MM-DD>-<meeting-slug>.md

• slug from title: lowercase, hyphens, max 40 chars
• Examples: 2026-03-31-1-1-aarav.md, 2026-03-31-slt-q2-planning.md
• Duplicate handling: append -2, -3 suffix

Never move earshot capture directories. Routing files the notes only; the capture directory (audio, transcript.*, narrative.md) stays where earshot wrote it. If the notes carry a **Recording**: <capture-dir> header (added by thinkkit's take-notes), preserve that line verbatim when moving or rewriting the notes — it stays valid precisely because the capture directory doesn't move.

Step 5 — Update SQLite index

Insert into file_index, per-folder index table (if exists for destination), search_fts with full note content.

If the meeting has an earshot capture with a narrative.md, index the narrative into search_fts as well — it is the richest artifact for the meeting. Index it at its in-place path inside the capture directory (it does not move).

Step 6 — Write to session-log.md

## YYYY-MM-DD HH:MM | pka-meetings | Filed: <title> → <path> | Action items: N | <open threads if any>

If a narrative.md was indexed, include its path in the entry: ... | Narrative: <capture-dir>/narrative.md.

If action items added to log:

## YYYY-MM-DD HH:MM | pka-meetings | Open: <action> | — | —

Step 7 — Confirm and summarize

Report: filed to path, attendees linked (N of M), action items found (list), transcript disposition.

Transcript Detection at Session Start

The PKA orchestrator's session-start scan flags transcript files in team-inbox/ and surfaces them:

"Found a transcript in team-inbox: GMT20260331-162935_Recording.transcript.vtt. Do you have notes to reconcile against it, or should I generate notes from the transcript directly?"

Unreconciled earshot captures are surfaced the same way:

"Found an earshot capture with an unreconciled transcript: meeting-notes/2026-07-02_140000_slt-sync/. Reconcile your notes against it?"

Key Constraints

• Transcript disposition always asked — never auto-deleted
• Earshot capture directories (marked by meeting.json) are never moved, renamed, or deleted — only the notes file routes
• **Recording**: header pointers in notes are preserved verbatim through routing and rewrites
• Reconcile (and narrate) before routing — notes must still be in meeting-notes/ for earshot's meeting-narrative to find them
• Project meeting routing always asks before routing to active project workspace
• Degrades gracefully when thinkkit skills are absent
• Never writes to project workspaces without asking