LegalQuants — Claude Code plugin

AI access to the LegalQuants community knowledge from your terminal — both the raw chat archive and a curated synthesis vault. Ask natural questions across everything, and — if you're a member — get a personalised experience that knows your contributions.

Members-only. You need a published LegalQuants profile (legalquants.com) to sign in.

---

Install

In Claude Code:

/plugin marketplace add https://github.com/LegalQuants/lq-plugin.git
/plugin install lq@legalquants

Use the full HTTPS URL above. The owner/repo shorthand clones over SSH and fails with Host key verification failed unless you've set up a GitHub SSH key — HTTPS needs no auth for this public repo.

Sign in (members)

/lq --signin

This starts a Google sign-in via a one-time device code:

1. The plugin shows a short code (e.g. ABCD-EFGH) and the link https://www.legalquants.com/device.
2. Open the link, enter the code, and sign in with the Google account on your LegalQuants profile.
3. The plugin caches a 7-day session locally (~/.config/lq/token.json, mode 0600) — no password or token to copy by hand.

Restart your Claude Code session so it picks up the sign-in, then:

/lq:start

If your profile is published and linked, you'll get an "I know you" greeting derived from your chat activity. Then just ask anything about the community.

/lq --signout clears your session.

---

What you get

• /lq:start (bare /lq is a kept alias) — cold-start interview + personalised orientation. Active members get an "I know you" greeting (no questions); quieter/guest members get a short interview. A plain question spans both corpora automatically — results are labelled by source.
• /lq:ask "<question>" — cross-source synthesis: fans out over both sources in parallel and merges into one cited answer. Use for "what's the community's take on X — and where's it from / what's the latest?"
• /lq:assess — assessment workflow (for invited candidates).
• One connector (lq-mcp) over the community chat archive + curated synthesis vault — read-only; one sign-in covers it. Tools take source: chat | brain | all (default all), so you don't pick a corpus up front; a plain query spans both and results are labelled by source.
• Auto-loaded guidance — one skill priming the model on the corpus's idioms.

What it does NOT do

• No writes — lq-mcp is read-only.
• No real-time ingest — chat is a sanitized snapshot; the brain vault is rebuilt periodically by operators.
• Never exposes another member's identity — /api/whoami returns only your own.

---

How sign-in works (under the hood)

Google login → your LegalQuants profile → a private builder ID → a 7-day Firebase session cookie. The cookie carries your identity as Firebase custom claims; the lq-mcp server verifies it keylessly (against Google's public certs — no service-account key) and returns only your own builder ID + first-name greeting. Your builder ID is never shown to other members.

Guests (no member sign-in) can still read the corpus via a shared bearer token, without personalisation.

Troubleshooting

• Commands don't appear after install — restart your Claude Code session (slash commands load at start).
• Sign-in rejected — your LegalQuants profile must be published. Publish it at legalquants.com, then /lq --signin again.
• MCP 401 after sign-in — restart the session so the cached cookie loads; or re-run /lq --signin if your 7-day session expired.

Privacy

The corpus is sanitized — authors appear as stable builder-NNN pseudo-IDs, never real names. Your cached session cookie is your credential on your machine; the server never reads your local files.

---

Questions: j.tso@legalquants.com · Built by LegalQuants.