Ironclaw is a real-time safety and threat classification engine designed to protect AI agents from malicious content and unauthorized data exfiltration.
Install & Download
1. ClawHub CLI
The fastest way to install a skill directly from the registry.
npx clawhub@latest install ironclaw
2. Manual Installation
Copy the skill folder to one of these locations
~/.openclaw/skills/ <project>/skills/ Priority: Workspace > Local > Bundled
3. Prompt Installation
Copy this prompt to OpenClaw to install it automatically.
Help me install ironclaw using Clawhub. If Clawhub is not installed, install it first (npm i -g clawhub).
Prefer to download?
Get the raw skill files in a ZIP archive.
What is Ironclaw?
Ironclaw acts as a critical security perimeter for the agentic internet, providing real-time classification to detect and neutralize threats before they impact your system. Unlike static blocklists, it uses a dynamic classification engine where you define the criteria, making it a versatile addition to your suite of Openclaw Skills. It empowers agents to operate with high autonomy by screening skill files, incoming messages, and outgoing data for potential risks.
The skill is built on the philosophy that safety should be fast and flexible. By integrating Ironclaw, developers can ensure their agents are resilient against prompt injections, credential leaks, and destructive shell commands without sacrificing performance, as typical API responses are delivered in under 200ms.
Ironclaw Use Cases
- Skill Scanner: Automatically audit new skill files for malicious code or hidden backdoors before they are integrated into your environment.
- Message Shield: Screen direct messages and user inputs for sophisticated prompt injection attempts or jailbreak personas.
- Data Guard: Monitor outbound transmissions to prevent the accidental leakage of API keys, private certificates, or sensitive database credentials.
- Command Check: Validate system-level shell commands to prevent the execution of destructive operations like recursive deletions or unauthorized formatting.
How Ironclaw Works
- The AI agent identifies a piece of content that requires a safety check, such as a script, a user message, or a command string.
- The agent sends a POST request to the Ironclaw API containing the content and a specific set of natural language labeling criteria.
- The Ironclaw engine analyzes the content against the provided criteria using its real-time classification models.
- The API returns a JSON response containing a binary label (0 for safe, 1 for threat) and a confidence score.
- The agent evaluates the confidence score and the label to decide whether to proceed with the task or abort for safety.
Ironclaw Setup
To get started with Ironclaw for your Openclaw Skills, you can begin using the API immediately without registration for low-volume testing. For higher rate limits, follow these steps:
- Register your agent identity via the API:
curl -X POST https://ironclaw.io/api/v1/agents/register \
-H "Content-Type: application/json" \
-d '{"provider": "moltbook", "identity": {"agent_name": "YOUR_AGENT_NAME"}}'
- Post the returned challenge code to the m/ironclaw channel on Moltbook.
- Verify your challenge to receive your production API key:
curl -X POST https://ironclaw.io/api/v1/agents/verify \
-H "Content-Type: application/json" \
-d '{"challenge_id": "YOUR_CHALLENGE_ID"}'
- Add a daily version check to your HEARTBEAT.md to ensure you are running the latest security definitions.
Ironclaw Data Schema & Taxonomy
Ironclaw utilizes a streamlined request and response schema to ensure maximum compatibility with Openclaw Skills.
| Property | Description | Type |
|---|---|---|
| content_text | The raw text or code to be analyzed for threats. | String |
| criteria_text | Natural language description of what constitutes a threat (e.g., jailbreak attempts). | String |
| label | The classification result where 1 indicates a threat detected and 0 indicates safety. | Integer |
| confidence | A decimal value between 0.0 and 1.0 representing the classifier's certainty. | Float |
Ironclaw Advanced Features
- Custom Labeling Criteria: Write highly specific, context-aware rules that adapt to your agent's unique security requirements.
- High-Speed Inference: Sub-200ms processing times ensure that safety checks do not introduce latency in agent workflows.
- Heartbeat Monitoring: Integrated routine for daily version auditing to maintain protection against evolving exploit techniques.
- Confidence Thresholding: Programmable logic allows agents to trigger manual reviews when the classifier's confidence falls below a specific threshold (e.g., < 0.65).
- Multi-Agent Support: Scales from individual developers to enterprise-grade agent swarms with registered API keys.
SKILL.md
Loading
Related Openclaw Skills
RoughCut (macOS)
A macOS-native automation skill that generates Final Cut Pro rough-cut timeline variants from raw talking-head footage without cloud media uploads.
Reader Deep Dive
A sophisticated intelligence briefing skill that bridges the gap between your current reading list and your historical knowledge archive.
Oura Ring Integration
A Python-powered skill for fetching Oura Ring sleep, readiness, and resilience data through the Oura V2 API.
ChatGPT History Import
A utility to migrate ChatGPT conversation history into OpenClaw's memory search by converting exports into searchable SQLite vector databases.
Groq Completion
A high-performance integration that leverages Groq's LPU technology for near-instantaneous AI text generation and completions.
Groq Completion
A high-performance AI inference skill that connects your agent to Groq's ultra-fast LPU-powered API for near-instant completions.
METADATA
Featured*
