Using Skills with OpenClaw

Overview

OpenClaw is an open-source personal AI assistant that runs entirely on your machine. Created by Peter Steinberger, it uses Claude or OpenAI as its backbone while keeping your data local. For lawyers working with sensitive documents --- client contracts, internal memos, privileged correspondence --- that distinction matters. Nothing leaves your machine unless you explicitly tell it to.

OpenClaw has a skills and plugins architecture that makes it a natural fit for Compound Skills. You place a Skill file in the right directory, point OpenClaw at a document, and the assistant applies the structured legal methodology the same way Claude or Codex would --- except everything happens locally.

This guide covers installation, setting up Skills, running analyses, and the features that make OpenClaw particularly useful for legal workflows.

Installing OpenClaw

OpenClaw installs from the command line with a single command:

curl -fsSL https://openclaw.ai/install.sh | bash

This sets up the core assistant. On macOS, there is also a companion app that provides a native interface --- menu bar access, drag-and-drop file handling, and system-level integration. Download it from the OpenClaw website after running the install script.

Once installed, configure your AI provider. OpenClaw supports both Anthropic (Claude) and OpenAI as backends. Add your API key in the settings, and the assistant is ready to use.

Setting Up Skills

OpenClaw organises its capabilities through a skills directory. To add a Compound Skill, download the .md file from the Compound Skills library and place it in the skills folder.

The default location depends on your setup, but a typical structure looks like this:

~/.openclaw/
├── config.yaml
├── skills/
│   ├── nda-review.md
│   ├── gdpr-data-request.md
│   ├── employment-reference-analysis.md
│   └── gmbh-formation-check.md
└── memory/

OpenClaw reads Markdown files in the skills directory automatically. There is no activation step or plugin installation. Once the file is in place, the assistant knows about it.

If you want OpenClaw to prioritise certain Skills or map them to document types, you can configure this in config.yaml. For example, you can tell OpenClaw to suggest the NDA Review skill whenever it detects a confidentiality agreement.

Running an Analysis

With a Skill in place, running an analysis follows a straightforward pattern.

Step 1 --- Load the document. Drag a file into the OpenClaw companion app, or reference it from the command line. OpenClaw handles PDF, DOCX, and plain text.

Step 2 --- Tell OpenClaw to apply the Skill. Be specific about which Skill to use. Example prompts:

• “Use the NDA Review skill to analyse this agreement.”
• “Apply the GDPR Data Request skill to this email.”
• “Run the Employment Reference skill on this Arbeitszeugnis and decode the hidden language.”
• “Check this Gesellschaftsvertrag with the GmbH Formation skill.”

Step 3 --- Review the structured output. OpenClaw works through the Skill’s methodology step by step, producing sections that correspond to each step in the Skill. Flagged issues, risk indicators, and statutory references appear in the order the Skill defines them.

Step 4 --- Ask follow-up questions. The initial pass surfaces the issues. The follow-up is where you dig in. Ask OpenClaw to elaborate on a specific clause, suggest alternative wording, or compare a provision against market standard.

OpenClaw-Specific Advantages

Several features make OpenClaw particularly well-suited for legal work with Compound Skills.

Local execution. Your documents never leave your machine. For law firms handling privileged or confidential material, this removes the data residency and confidentiality concerns that come with cloud-based tools. The AI model processes your queries via API, but document content stays local.

Persistent memory across sessions. OpenClaw maintains context between conversations. If you analysed an NDA last week and now need to review the corresponding service agreement, OpenClaw can recall the earlier analysis. This is especially valuable for ongoing matters where multiple documents relate to the same transaction.

Messaging integrations. OpenClaw can connect to WhatsApp, Telegram, and Slack. Once an analysis is complete, you can route the results directly to a channel or chat --- useful for sending a quick summary to a colleague or flagging an urgent issue without switching applications.

Browser control for research. OpenClaw can browse the web on your behalf. If a Skill flags a statute or court decision, you can ask OpenClaw to look up the current version or check for recent amendments. This pairs well with Skills that reference evolving regulatory frameworks like the AI Act or DSGVO enforcement decisions.

Best Practices

The same principles that apply to using Skills with Claude and Codex apply here.

Be explicit about which Skill to use. Naming the Skill ensures OpenClaw follows the defined methodology rather than generating a generic review.

Provide the full document. Partial uploads lead to partial analyses. The Skill needs the complete text to check for missing provisions and clause interactions.

Chain Skills for complex matters. Run the NDA Review skill first, then apply the Data Processing Agreement skill to the DPA that accompanies it. Tell OpenClaw when you are switching: “Now apply the DPA Review skill to this Auftragsverarbeitungsvertrag.”

Leverage persistent memory. Unlike stateless tools, OpenClaw remembers prior sessions. Reference earlier analyses explicitly: “Compare this NDA to the one we reviewed last Tuesday for the same client.”

Verify everything. Skills structure the analysis. You make the call. Check statutory references, confirm that cited provisions are current, and apply the professional judgment that only a practising lawyer can bring. AI output is a structured first draft --- not a legal opinion.