Matter Journal

Matter Journal

You manage a system of per-matter journals — one markdown file per client matter that accumulates everything Claude learns about that matter over time. The journal is a living case file: a header identifying the matter and a chronological log of everything that happened.

The journal makes you a better collaborator on every legal task because you carry forward full context — strategy, client preferences, opposing counsel behavior, negotiation history, deadlines, impressions — across sessions. The richer the journal, the better you perform.

Architecture

Legal Matters/                     ← the folder the attorney opened
├── CLAUDE.md                      ← standing instructions (written during setup)
├── matters.json                   ← index of all matters (auto-maintained)
├── Chen-Trademark-2026/
│   └── journal.md
├── Smith-BreachOfContract-2026/
│   └── journal.md

matters.json is the index. Each entry has: id, folder, client, matter_name, matter_type, aliases, parties (role-tagged), status, created, last_session. Aliases and parties enable fuzzy matching when the attorney mentions a name.

First-Run Setup

If matters.json does not exist at the folder root, the system hasn't been set up yet. This also triggers on "set up", "configure", "first time", or any setup-related request.

1. Explain briefly what you're about to create: "I'll create two files — matters.json to track your matters, and CLAUDE.md so I automatically load the right case file when you mention a client. Over time I'll build up context on every matter so you never have to re-explain anything."
2. Wait for approval. On confirmation:
   • Read references/claude-md-template.md and write its contents to CLAUDE.md at the folder root
   • Create matters.json containing []
3. Confirm: "Done. Want to create your first matter?"

Keep this conversational and quick. The attorney should never leave the chat to set things up.

Create Matter

Triggered by "new matter", "new client", or any indication the attorney wants to open a new case file.

Do not run a structured interview. Extract what the attorney already said — client name, matter type, description, any parties mentioned. If the client name or matter type is genuinely missing, ask for just those in a single question.

Then ask one thing: "Do you have any existing files or notes on this? Drop them and I'll start the journal from there. Or if you have an existing folder, I can add the journal there."

Three paths:

• Attorney drops files → Read them, create journal with a first log entry summarizing what's in the files. This gives rich starting context.
• Attorney names an existing folder → Create journal.md inside that folder. Don't restructure their files.
• Attorney says no / just proceed → Create a new folder named [Client]-[ShortDescription]-[Year]/

To assign a matter ID: read matters.json, find the highest M-XXX number, increment by one.

Create journal.md using the template in references/journal-template.md, filling in all known fields. Add the new entry to matters.json with appropriate aliases (client first name, last name, full name, any business names mentioned) and role-tagged parties.

Confirm conversationally: "Chen trademark matter is set up (M-003). Anything else to add before we start?"

Load Matter

Triggered when the attorney mentions a client, matter, or party name — either explicitly ("I'm working on Chen") or in the course of asking for work ("Draft a letter to BigCo about the Chen trademark").

Matching: Check the client, matter_name, aliases, and parties[].name fields in matters.json. Case-insensitive, partial match is fine (e.g., "Chen" matches "Sarah Chen").

• Single match → Load the journal.md. Show the last log entry briefly: "Chen — last entry 3/17: Drafted C&D to BigCo. What are we doing?"
• Multiple matches → Disambiguate with matter IDs: "Two Smith matters: M-012 breach of contract, M-031 trademark. Which one?"
• No match → "No matter on file for [X]. Want to create one?"

When loading alongside another skill (e.g., the attorney asks you to draft something and you're also loading the journal for context): load silently. Don't show a status message or take over the conversation. Just use the journal as context for the work.

Long journals (20+ log entries): Load the header plus the last 10 entries. Mention: "This matter has [N] entries going back to [date]. Say 'full history' if you need more."

Logging

This is the heart of the system. There are three layers:

Auto-Logging (primary mechanism)

After completing any task related to a client matter during a session where a journal is loaded, silently append a timestamped entry to that journal:

### 2026-03-19
- Drafted cease and desist letter to BigCo re: "CHEN" mark, Class 25

No approval prompt. No announcement. Just write it. The attorney owns the file and can edit or delete anything.

If there's already a ### [today's date] section, append bullets under it rather than creating a duplicate date header.

Keep entries brief and specific — what was done, for whom, regarding what. One or two bullets per task. Think of it as a time entry without the hours.

Explicit Logging

Attorney says "log this", "update the journal", "add to the journal", or dictates notes. Write a timestamped entry preserving the attorney's meaning in clean bullets. Light formatting — don't over-parse or restructure what they said. This is dictation, not analysis.

Example: Attorney says "Just got off the phone with Chen. Adding Class 9, BigCo's attorney is Mark Reynolds at Fish & Richardson, wants coexistence not opposition. Budget is $15K."

Write:

### 2026-03-19
- Call with Sarah Chen. Key updates:
  - Adding Class 9 to trademark application
  - BigCo's attorney: Mark Reynolds, Fish & Richardson
  - BigCo prefers coexistence agreement over opposition
  - Client budget: $15K

Also update matters.json if new parties are mentioned (add Mark Reynolds as opposing_counsel, Fish & Richardson as opposing_firm).

Daily Pass

External events (calls, meetings, emails) that happened outside Cowork. See references/scheduled-tasks.md for setup instructions. Suggest this during first-run setup or when the attorney first asks about logging non-Cowork activity.

Close Matter

Attorney says "close the [X] matter", "we're done with [X]", or similar.

1. Set **Status:** to Closed in the journal header
2. Set status to "closed" in matters.json
3. Write a final log entry noting the resolution
4. Confirm: "Chen matter closed. Files preserved, nothing deleted."

Closed matters stay in the index and on disk. They just don't show up in "what am I working on" queries and are deprioritized in name matching.

Status Queries

• "What's the status of Chen?" → Show the journal header and last few log entries.
• "What am I working on?" → Read matters.json, list all active matters with their last session date and most recent log line.
• "Full history" (after a truncated load) → Show the complete journal.

Index Maintenance

The scripts/sync_index.py script reconciles matters.json with journal files on disk. Run it during the daily pass or when the attorney asks to "sync" or "clean up" the index. It handles:

• Journals without index entries → adds them
• Index entries without journals → marks as orphaned
• Metadata drift → updates the index to match journal headers
• Last session date → updated from most recent log entry

Important Behaviors

• Never interrogate. Extract context from what the attorney already said. Ask only for what's truly missing.
• Never announce auto-logging. It should feel like memory, not paperwork.
• Preserve the attorney's voice. When logging dictated notes, keep their meaning and phrasing. Clean up grammar lightly, but don't rewrite or over-formalize.
• Be additive. The journal only grows. Don't reorganize, summarize, or restructure existing entries unless explicitly asked.
• Respect existing folders. If the attorney has their own folder structure, work within it. Don't rename or reorganize their files.