sboghossian/master-claude-for-legal

> A community skill pack for legal teams. Set up Claude correctly for privileged work, customize the legal plugin without faceplanting, and ship your first production legal automation in an afternoon.
> Maintained by HAQQ Legal AI. MIT licensed. Contributions welcome.
## What this is
A self-contained skill pack you can install into Claude (Cowork, Claude Code, or any tool that supports Anthropic-format skills) that teaches the model — and you — how to do legal work with Claude the way Anthropic's own legal team does it. Built from both of Anthropic's public legal webinars in 2026: Claude for Legal Teams (April) and How Legal Teams Put Claude to Work (May).
It comes in four layers.
Reference docs (references/). Fifteen short documents covering the four-pillar architecture, the four privilege layers, the vocabulary that confused the room at the April webinar, MCP permission hardening, the May connector catalog (Box / Harvey / Thomson Reuters / Free Law Project / Courtroom5 / etc.), the 12 practice-area plugins, the Microsoft 365 cross-surface context model (Word / Excel / PowerPoint / Outlook), managed agents, the cold-start interview customization ritual, citation verification, the agentic harness pattern for long documents, practice-area patterns, the four-hour setup checklist, a skill-authoring guide, and a list of anti-patterns we've seen blow up in production.
Starter skills (skills/). Six working skills you can install today: NDA triage, multi-party version diff, tabular review (the May webinar's demoed M&A diligence pattern), meeting brief, citation verifier, status synthesis (the "Friday newsletter" pattern). Each is a markdown file. Each is meant to be remixed for your firm — the legal plugins Anthropic ships are starting templates, not finished products, and so is this.
Templates (templates/). Three documents most firms need but few have written: a firm AI policy template, a one-page client-facing explainer for "how does our firm handle your data when using AI," and a vendor security questionnaire for assessing legal-AI tools.
Data (data/). Full transcripts of both source webinars, the 51 audience questions from the April session with upvotes and themes, and machine-readable JSON for anyone who wants to do their own analysis. The directory is the raw output of the cleanup-and-structuring effort; everything else in the repo is downstream of it.
## Why this exists
In April 2026, Anthropic ran a webinar called Claude for Legal Teams. Twenty thousand lawyers registered. The chat submitted 51 questions and upvoted them 2,470 times. About half got answered live; the other half didn't, and several of the answers that did happen were partial.
In May 2026, Anthropic ran a follow-up webinar — How Legal Teams Put Claude to Work — announcing the "connected legal stack": 12 practice-area plugins, 20+ MCP connectors, the Microsoft 365 cross-surface integration, and managed agents. Mark Pike (Anthropic Associate General Counsel and product lead for Claude for Legal) presented again, this time with Harry from the Applied AI team. The May session went deep on what changed and what to do with it.