ctx-transcript

name	ctx-transcript
description	Process a new stakeholder meeting transcript. Collects inputs interactively (project, transcript, attendees, date), then runs synthesis, validation, and shareable output in the background so you can keep working.
user_invocable	true

ctx-transcript

Two phases:

1. Interactive: Confirm project, get transcript, confirm attendees and date
2. Background: Synthesis → validation → metadata → shareable output

Supports --batch flag for processing multiple transcripts at once.

---

Phase 1 — Collect Inputs

Step 1 — Print:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  ADD A TRANSCRIPT
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Then print this one-line tip (non-blocking — proceed to Step 1b without waiting):

Tip: /add-context auto-detects type — you can route through it next time instead of /ctx-transcript.

Step 1b — Batch mode check.

Read [workspace-root]/.current-session to identify the active project and its PROJECT_ROOT.

If --batch flag passed: skip to Batch Mode section below.

Otherwise, check [PROJECT_ROOT]/intelligence/meetings/ for any subdirectories containing a synthesis.md file (i.e. processed meetings).

• No processed meetings found → print:

  No meetings processed yet for this project.
  
  Want to load several transcripts at once? Drop your files into:
    [PROJECT_ROOT]/intelligence/meetings/inbox/
  
  Then press Enter and I'll process them all together.
  Or paste a single transcript now to get started.
  
  [batch / paste]
  

  Wait for response. If batch: go to Batch Mode. If paste: continue to Step 2.

• Processed meetings exist → continue to Step 2 (normal flow).

---

Batch Mode

B1 — Create inbox. Create [PROJECT_ROOT]/intelligence/meetings/inbox/ if it doesn't exist.

B2 — Wait for files. If the user selected batch mode, print:

Ready. Drop your transcript files (.md or .txt) into:
  [PROJECT_ROOT]/intelligence/meetings/inbox/

Press Enter when ready.

Wait for confirmation.

B3 — Read inbox. List all files in inbox/. If empty: print No files found in inbox. Add files and try again. and stop.

B4 — Prepare each file. For each file in the inbox:

1. Read the first 40 lines to extract: date, time, attendee names, topic.
2. Match attendees against root people/ and [PROJECT_ROOT]/people/. Note any unmatched names.
3. Build MEETING_FOLDER: [PROJECT_ROOT]/intelligence/meetings/[DATE]-[TIME]-[slug]/
4. Copy file content verbatim to [MEETING_FOLDER]/raw.md.
5. Delete the source file from inbox/.

Hold the prepared meeting params (folder path, raw file path, attendee map, date) in a list — they all dispatch together in B5.

B5 — Dispatch agents in parallel and wait. Spawn one Agent call per prepared meeting in a single message so they run concurrently. Each agent gets the same parameters as the single-transcript handoff and the instruction: Read .claude/skills/ctx-transcript/BACKGROUND.md for processing instructions. Do not ask questions — proceed through all steps.

Use run_in_background: false (the default) so the main thread waits for all agents to complete. Sync wait is what makes the verification in B6 possible — without it, B6 would run while agents were still mid-write and report false failures.

B6 — Verify outputs and report. For each meeting folder created in B4, check that the expected output files exist and are non-empty:

• [MEETING_FOLDER]/synthesis.md — must exist, must be > 0 bytes
• [MEETING_FOLDER]/validation.md — must exist, must be > 0 bytes
• [MEETING_FOLDER]/shareable.md — must exist, must be > 0 bytes

If any file is missing or 0 bytes, that meeting's processing failed. The most likely cause: auto-accept edits was off and the agent's writes got silently denied (background-style permission requests can't surface to the user).

If any failures, print:

⚠ Output verification failed for [N]/[M] meetings:
  - [meeting-folder]: synthesis.md is 0 bytes
  - [meeting-folder]: shareable.md missing

Most likely cause: auto-accept edits was off during processing.

Fix: turn on auto-accept (Shift + Tab until your status bar shows auto-accept or auto mode on), then re-run:
  /ctx-transcript --batch

If all files verify clean, print:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  BATCH COMPLETE
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
Processed: [N] transcripts · all outputs verified

⚠ Gaps to resolve:
- [filename]: date not found — assumed [date] from filename
- [filename]: "Alex M" unmatched — no contact record found
- [filename]: attendees unclear — only one speaker detected

If there are no gaps, omit the warning block. Syntheses are written; the user can run /ctx-synthesise next.

---

Phase 1 — Single Transcript (normal flow)

Step 2 — Project. Read [workspace-root]/.current-session for the active project. Show pre-selected, list others from projects.md. Wait for confirmation.

Step 3 — Transcript. Ask: Paste the transcript below, or give a file path:. Accept either. If a path: read only the first 30 lines (header block) to extract attendees and date/time — do not read the full file. Pass the path to the background agent.

Step 4 — Attendees. Identify speakers from header block or first few lines. Match each against BOTH root people/ AND [PROJECT_ROOT]/people/ by first name, full name, or role keyword. Show confirmation table — full name, role, matched/unmatched.

For unmatched names, offer to create a contact: extract name, role, company from what's visible and write:

• Root stub: people/[name-slug].md (using people/TEMPLATE.md)
• Project profile: [PROJECT_ROOT]/people/[name-slug].md (see File Structure below)

For matched names who don't yet have a project-level profile, create [PROJECT_ROOT]/people/[name-slug].md automatically.

After creating any new project-level profile, update the root people/[name-slug].md Project Footprint table. Append a row | [Project Name] | [Role from project profile] | if not already present (idempotent — skip if the project is already listed). Project Name = the name from projects.md. Role = the **Role:** value from the project profile.

Skip "Me" / "You" / the user.

Step 5 — Date/time. Check transcript headers first. If found, confirm with user (Y/N). If not found, ask. Get duration too.

---

Handoff to Background Agent

Say: Got everything. Processing in the background — you can keep working.

Spawn background agent (run_in_background: true). Pass:

• PROJECT_ROOT — e.g. projects/acme-corp-launch/
• TRANSCRIPT_PATH — file path if given; if pasted, write verbatim to [MEETING_FOLDER]/raw.md first, then pass that path
• ATTENDEES — list of {full_name, role, company, people_file_path}
• MENTIONED — names referenced but not attending
• DATE, TIME, DURATION
• MEETING_FOLDER — [PROJECT_ROOT]/intelligence/meetings/[DATE]-[TIME]-[attendee-slug]/

Slug = attendee names joined by hyphens (exclude user), max 40 chars.

Tell the agent: Read .claude/skills/ctx-transcript/BACKGROUND.md for processing instructions. Do not ask questions — proceed through all steps.

---

File structure

[PROJECT_ROOT]/intelligence/meetings/[DATE]-[TIME]-[slug]/
  raw.md          ← verbatim transcript, never modified
  synthesis.md    ← validated synthesis + YAML frontmatter
  validation.md   ← per-statement evidence audit
  shareable.md    ← professional summary, safe to send to attendees

[PROJECT_ROOT]/people/[name-slug].md   ← project-level profile (created/updated per attendee)
people/[name-slug].md                  ← root-level identity stub (created if new contact)

Project-level people file format ([PROJECT_ROOT]/people/[name-slug].md):

# [Full Name] — [Project Name]

**Role:** [Title, Company]
**See also:** [Root profile](../../people/[name-slug].md)

---

## Their Role Here

[Fill in after first meeting]

---

## What They Care About

[Fill in after first meeting]

---

## Working With Them

[Fill in after first meeting]

---

## Key Context

[Fill in after first meeting]

---

## Sources

The background agent populates Their Role Here from the transcript if clear. Other sections are left as placeholders for the user to fill in, or for ctx-doc to populate from richer material.