The Legal Agency

An open-source, multi-agent framework for AI-native legal practice.

The Legal Agency is a collection of 30 specialized AI agent definitions organized across 3 divisions, designed to power a comprehensive AI-native legal firm. Each agent is a self-contained markdown file with structured identity, workflows, deliverables, and guardrails — ready for integration into any multi-agent orchestration platform.

npm install @judicialmind/legal-agents

Project Structure

legal agents/
├── README.md
├── ROSTER.md                          # Master index of all 30 agents
│
├── practice-verticals/                # Division 1 — Core legal practice areas
│   ├── corporate-ma.md
│   ├── litigation.md
│   ├── intellectual-property.md
│   ├── real-estate.md
│   ├── employment-labor.md
│   ├── tax-law.md
│   ├── immigration.md
│   ├── criminal-defense.md
│   ├── family-law.md
│   ├── bankruptcy.md
│   ├── healthcare-law.md
│   └── environmental-law.md
│
├── functional-domains/                # Division 2 — Cross-cutting legal operations
│   ├── legal-research.md
│   ├── contract-lifecycle.md
│   ├── e-discovery.md
│   ├── due-diligence.md
│   ├── compliance-tracking.md
│   ├── case-management.md
│   ├── client-intake.md
│   ├── legal-analytics.md
│   ├── billing-operations.md
│   └── document-drafting.md
│
└── jurisdictions-specialties/         # Division 3 — Jurisdiction-specific & specialist skills
    ├── us-federal-courts.md
    ├── india-legal.md
    ├── uk-commonwealth.md
    ├── international-arbitration.md
    ├── regulatory-compliance.md
    ├── court-filings.md
    ├── deposition-prep.md
    └── brief-writing.md

Divisions

Division 1 — Practice Verticals (12 Agents)

Domain-specific agents covering the core areas of legal practice. Each agent embodies a senior specialist in its field with deep knowledge of applicable statutes, regulations, and case law.

Division 2 — Functional Domains (10 Agents)

Cross-cutting operational agents that support any practice area with specialized workflows for research, document management, billing, analytics, and more.

Division 3 — Jurisdictions & Specialties (8 Agents)

Jurisdiction-specific experts and specialist skill agents for procedural tasks that cut across practice areas.

Agent File Format

Every agent definition file follows a consistent structure:

---
YAML Frontmatter
  name, category, emoji, vibe, services, version, tools
---

# Agent Name
Introduction paragraph

## 🧠 Identity & Personality Traits
## 🎯 Core Mission & Workflows
  - Mission statement
  - 3-Level Progressive Disclosure (Trigger → Skill Body → References)
  - Standard Operating Procedure (SOP)
## 📋 Technical Deliverables & Examples
  - 3–4 concrete deliverables with formatted templates
## 🎯 Success Metrics & Guardrails
  - Measurable success criteria
  - Zero Hallucination Tolerance rules

Key Design Principles

• 3-Level Progressive Disclosure: Agents activate on trigger keywords (Level 1), apply domain-specific methodology (Level 2), and reference authoritative sources (Level 3) — minimizing token usage while maximizing depth
• Concrete Deliverables: Every agent includes structured templates with realistic examples (tables, checklists, formatted documents) — not abstract descriptions
• Zero Hallucination Guardrails: Each agent explicitly forbids fabrication of citations, rule numbers, case law, financial figures, and other domain-specific data points

Installation

npm install @judicialmind/legal-agents

Programmatic Usage (Node.js)

const { getAgent, listAgents, listDivisions, searchAgents } = require('@judicialmind/legal-agents');

// Get a single agent by slug
const agent = getAgent('corporate-ma');
console.log(agent.name);      // "Corporate M&A Counsel"
console.log(agent.services);  // ["M&A transaction advisory", ...]
console.log(agent.body);      // Full markdown body

// List all agents in a division
const litigators = listAgents({ division: 'practice-verticals' });
console.log(litigators.length); // 12

// List every agent across all divisions
const all = listAgents();
console.log(all.length); // 30

// Get available divisions
const divs = listDivisions();
// ["practice-verticals", "functional-domains", "jurisdictions-specialties"]

// Search agents by keyword
const results = searchAgents('arbitration');
console.log(results.map(a => a.name));
// ["International Arbitration Counsel"]

Agent Object Shape

interface Agent {
  id: string;         // file slug, e.g. "corporate-ma"
  file: string;       // relative path, e.g. "practice-verticals/corporate-ma.md"
  name: string;       // display name
  category: string;   // division slug
  emoji: string;      // icon
  vibe: string;       // one-line personality
  services: string[]; // capabilities
  version: string;    // semver
  tools: string[];    // allowed tool names
  body: string;       // markdown body (after frontmatter)
  raw: string;        // full raw file content
}

Direct Usage (without npm)

These agent definitions are platform-agnostic markdown files. They can also be used directly:

• Multi-agent orchestration frameworks — load agent files as system prompts for specialized LLM instances
• Copilot / VS Code agent mode — use as .agent.md files for custom agent personas
• RAG pipelines — index agent deliverable templates as retrieval targets for document generation
• Prompt libraries — reference individual SOPs and deliverable templates as structured prompts

Security

This repository includes automated package hardening checks:

• npm run audit — runs npm advisory audit
• npm run security:scan-secrets — scans for common leaked credential patterns
• npm run security:pack-check — verifies forbidden files do not enter npm tarball
• npm run security:all — runs all package safety checks

GitHub Actions workflows:

• .github/workflows/security.yml runs on every push and pull request
• .github/workflows/release-check.yml validates release readiness and dry-run publish with provenance

Recommended release posture:

• Use npm Trusted Publishing (OIDC) for production releases
• Avoid long-lived npm tokens whenever possible
• Rotate any token immediately if it is ever exposed

License

MIT — see LICENSE for details.