ctx-chat

name	ctx-chat
description	Capture a WhatsApp, Slack, or DM thread into project context. Handles multi-party threads with deduplication — import the same thread repeatedly and only new messages get processed.
user_invocable	true

CTX Chat

Two phases:

1. Interactive: Confirm project, get thread content, confirm participants and thread name, handle deduplication
2. Background: Synthesise → validate → update thread index → write outputs

---

Phase 1 — Collect Inputs

Step 1 — Print:

━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━
  CAPTURE A CHAT THREAD
━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━

Then print these one-line tips (non-blocking — proceed to Step 2 without waiting):

Tip: make sure auto-accept edits is ON (Shift + Tab) — background synthesis fails silently otherwise.
Tip: /add-context auto-detects type — you can route through it next time instead of /ctx-chat.

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

Step 3 — Content. Ask: Paste thread below, or give a file path:

Step 4 — Participants. Extract participant names from speaker labels (Name:, [timestamp] Name:, bold patterns).

• Match each against [PROJECT_ROOT]/people/ by first name, full name, or role
• Show confirmation table:

Name found in content  →  Matched to               Action
─────────────────────────────────────────────────────────
Alex Chen              →  alex-chen.md              ✓ matched
Priya Sharma           →  priya-sharma.md           ✓ matched
Marcus Webb            →  No match found            Create stub?

Offer to create stubs for unmatched names.

Step 5 — Thread name. Ask: Thread name (used for deduplication): Suggest a slug based on participants or platform (e.g. acme-launch-team-whatsapp). User can edit.

Step 6 — Deduplication check. Check if [PROJECT_ROOT]/intelligence/chats/[thread-slug]/index.md exists.

• If yes: show last processed date range. Ask: Last import covered [date range]. Process from [end date] onwards? [Y/N]
• If no: first import — process all.

Step 7 — Context (optional). Ask: Any context? e.g. post-event, planning thread (type "none" to skip):

Step 8 — Date range. Check content for timestamps. If found, confirm. If not: Date or date range (e.g. 10–17 Apr 2026):

Show summary:

Project:      [name]
Thread:       [thread-slug]
Participants: [names]
Import:       [N] (new / existing thread)
Date range:   [range]
Folder:       chats/[thread-slug]/[date-range]/

Ask: Looks right? [Y/N]

---

Handoff to Background Agent

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

Spawn background agent (run_in_background: true). Pass:

• PROJECT_ROOT
• CONTENT_PATH — file path if given; if pasted, write verbatim to a temp path and pass that
• THREAD_SLUG
• IS_NEW_THREAD — true/false
• PROCESS_FROM_DATE — for existing threads, only synthesise content after this date
• PARTICIPANTS — [{full_name, role, people_file_path}]
• CONTEXT — optional context string
• DATE_RANGE — e.g. "2026-04-10_2026-04-17"
• CHAT_FOLDER — [PROJECT_ROOT]/intelligence/chats/[thread-slug]/[DATE_RANGE]/

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

---

File structure

[PROJECT_ROOT]/intelligence/chats/

  [thread-slug]/
    index.md                    ← thread identity + import history (deduplication record)
    [DATE-RANGE]/               ← e.g. 2026-03-30_2026-04-16
      raw.md
      synthesis.md
      validation.md