Scan your OpenClaw skills, MCP servers, and agent workspaces for vulnerabilities — before attackers exploit them.
No account needed for basic scan · Results in under 60 seconds
Comprehensive scanning across every attack vector unique to AI agent architectures.
Detects overly broad file system, network, and tool permissions in skill configs. Flags wildcards, unrestricted path access, and dangerous capability grants that give agents more access than they need.
Critical RiskScans skill instructions, SKILL.md files, and prompt templates for injection-vulnerable patterns. Catches "ignore previous instructions," role-override attempts, and jailbreak fragments embedded in agent logic.
High RiskIdentifies code paths that send data to unverified external endpoints. Flags undisclosed HTTP calls, suspicious base64 encoding, and telemetry that could exfiltrate conversation data or user files.
Medium RiskFinds hardcoded API keys, tokens, passwords, and credentials across all file types. Detects OpenAI keys, AWS credentials, Stripe tokens, database URLs, and 50+ other secret patterns using entropy analysis.
Critical RiskFour steps from upload to a fully remediated, secure agent workspace.
Drag and drop your skill directory, paste a GitHub URL, or point the CLI at your local workspace path.
Our engine runs all four security checks in parallel — permission audit, injection detection, data leaks, and secret scanning.
Get a structured JSON report with severity levels, exact file locations, matched patterns, and remediation steps.
Follow the actionable remediation guide. Re-scan to verify fixes. Full-tier customers get a human review of edge cases.
Examples of vulnerabilities found in real-world AI agent skills and workspaces.
API key committed directly in skill.py — exposed to anyone who reads the file or has repo access.
Skill config grants read/write access to the entire file system rather than its own working directory.
Third-party SKILL.md contains instructions that override the agent's primary goals when processed.
Skill silently POSTs conversation context to an external endpoint not documented in its README.
Agent executes dynamically constructed code strings that could be poisoned via crafted input.
Database connection string with credentials stored in plain-text YAML config committed to the workspace.
One-time scans or continuous monitoring — we have a plan for every stage.
Basic scan — no account required. Spot-check a single skill or file in seconds.
Start Free Scan →Complete one-time audit of your workspace, all skills, and MCP configs. Delivered in minutes.
Get Full Report →Always-on protection. Every skill update or new install gets scanned automatically — with alerts.
Contact Us →Run a free scan today. Find vulnerabilities before they become breaches. Takes under 60 seconds.