This directory is a complete, drop-in NemoClaw/OpenClaw configuration for the Legal solution segment. It defines one parent orchestration agent and twelve specialist subagents covering the full operational surface of a law firm.

Copy the contents of this directory into /sandbox/.openclaw/ on a NemoClaw sandbox and the configuration is ready to load — no conversion required.

The Firm Assistant (legal-assistant) is the default entry point. When an attorney, paralegal, or staff member sends a message, the Firm Assistant:

1. Identifies the intent (research, drafting, intake, billing, discovery, compliance, knowledge, conflicts, training, docketing, or IT).
2. Spawns the appropriate specialist subagent via sessions_spawn.
3. Waits for the subagent result.
4. Synthesizes and returns a clear, actionable response.

For multi-domain requests (e.g., opening a new matter requires a conflicts check before intake), the Firm Assistant orchestrates subagents in the correct sequence. Conflicts check always runs before client intake.

Follow these steps to deploy this configuration on a NemoClaw sandbox.

1. Copy the directory contents.

git clone https://github.com/iblai/legal-agents.git
cp -r legal-agents/. /sandbox/.openclaw/
cp legal/.env.example /sandbox/.openclaw/

Ensure the directory structure is preserved. The result should be:

/sandbox/.openclaw/
├── openclaw.json
├── .config-hash
├── .env.example
├── workspace/
│   └── .gitkeep
├── skills/
│   ├── clio/SKILL.md
│   ├── docket-alarm/SKILL.md
│   ├── docusign/SKILL.md
│   ├── imanage/SKILL.md
│   ├── intapp-conflicts/SKILL.md
│   ├── ironclad/SKILL.md
│   ├── netdocuments/SKILL.md
│   ├── pacer/SKILL.md
│   ├── relativity/SKILL.md
│   ├── servicenow/SKILL.md
│   └── westlaw/SKILL.md
└── agents/
    ├── legal-assistant/agent/
    ├── case-research-agent/agent/
    ├── contract-review-agent/agent/
    ├── client-intake-agent/agent/
    ├── billing-time-agent/agent/
    ├── discovery-agent/agent/
    ├── compliance-agent/agent/
    ├── knowledge-agent/agent/
    ├── brief-drafting-agent/agent/
    ├── conflicts-check-agent/agent/
    ├── training-agent/agent/
    ├── docket-management-agent/agent/
    └── it-help-desk-agent/agent/

Each agent directory contains the following workspace files:

2. Set ownership.

chown -R sandbox:sandbox /sandbox/.openclaw/

3. Set up credentials.

Two credential sources must be populated before the agents can call external platforms.

Inference credentials — every auth-profiles.json contains a non-functional placeholder API key. Replace the apiKey value in each file with a real Anthropic API key (or the appropriate credential for your inference provider):

find /sandbox/.openclaw/agents -name auth-profiles.json

Edit each file and replace the value of "apiKey" under profiles.anthropic.

Tool credentials — .env.example is a committed sample template representing the OpenClaw daemon env file ~/.openclaw/.env. It contains non-functional placeholder values for every third-party platform used by the segment. Copy it and fill in real credentials:

cp /sandbox/.openclaw/.env.example ~/.openclaw/.env
# then edit ~/.openclaw/.env and replace every SAMPLE-... value

.env.example is safe to commit; ~/.openclaw/.env (with real keys) must never be committed.

4. Recompute the config hash.

After any edit to openclaw.json (including none), recompute the hash:

cd /sandbox/.openclaw
sha256sum openclaw.json > .config-hash

On macOS use shasum -a 256 instead of sha256sum.

5. Reload the OpenClaw gateway.

openclaw reload
# or restart the NemoClaw service via your host's service manager

The skills/ directory contains one subdirectory per third-party platform, each containing a SKILL.md file. Each skill describes the platform's capabilities, the specific API operations, authentication patterns, and declares required env vars in its frontmatter metadata key. Agents load the relevant skill files at runtime to understand how to call external systems.

All skills draw credentials from ~/.openclaw/.env on the daemon host. The committed sample template .env.example contains non-functional placeholder values for every platform; copy it to ~/.openclaw/.env and replace each placeholder with a real credential before deploying (see Import Instructions, step 3).

• The parent agent (legal-assistant) is marked "default": true and will be the first agent users interact with.
• All agents use anthropic/claude-sonnet-4-5-20250929 by default. To change the model, update openclaw.json and recompute the config hash.
• The shared workspace at /sandbox/.openclaw/workspace/ is writable by all agents and is used to pass structured context between subagents within a session.
• Agents do not give legal advice. They assist qualified legal professionals. All outputs requiring professional judgment must be reviewed and approved by a licensed attorney.

Released under the MIT License.