ClawSec
Security skills for OpenClaw, NanoClaw, Hermes and Picoclaw: signed advisories, guarded installs and drift detection
Install
npx skills add prompt-security/clawsec --skill clawsec-suite -a openclaw --global -yFor OpenClaw. The advisory hook is enabled separately via setup_advisory_hook.mjs after reviewing its preflight.
This is third-party code. Review the repository files before installing.
What it does
ClawSec by Prompt Security protects Claw-family autonomous agent runtimes. The core clawsec-suite verifies a signed advisory feed of NVD CVEs and approved community reports and matches it against installed skills. Installing a skill with a known issue stops and requires a second confirmation. Separate packages watch critical agent files such as SOUL.md for drift, audit configuration, produce attestations and verify release signatures. Picoclaw also gets an opt-in self pen-testing package.
Who it is for. For operators running OpenClaw, Hermes, NanoClaw or Picoclaw who are responsible for their security.
Good fit when
- The agent installs skills from catalogs and you need pre-install checks
- You need to notice changes to the agent's identity or config files
- You need regular audits of the agent environment
Not a fit when
- You use Claude Code, Cursor or Codex without Claw platforms
- You need a network proxy: traffic-guardian packages are only specifications for now
Example request
Check installed skills against the ClawSec advisory feed and show everything high or criticalLimitations
Works only with OpenClaw, NanoClaw, Hermes and Picoclaw. The AGPL-3.0 license has obligations when distributing changes. The advisory hook changes persistent OpenClaw config and needs a gateway restart. Some workflows need WSL or Git Bash on Windows.
How to disable. Disable the advisory hook in OpenClaw config, delete clawsec-suite from ~/.openclaw/skills and restart the gateway.
Security check
- Changes persistent agent config and adds hooks
- Runs scripts and can set up scheduled jobs
- The self pen-testing package runs offensive checks
README in short
The README describes installing the OpenClaw suite in two separate steps: adding the package and enabling the hook after reviewing its preflight. It shows a drift detection demo, protection layers and entry points for each platform. It documents the signed advisory feed and how to query it with curl. AGPL-3.0 licensed.
FAQ
Does ClawSec remove skills by itself?
No, it recommends and gates risky actions; removal and overrides stay under human approval.
Where do advisories come from?
From NVD CVEs, approved community reports and provisional GitHub advisories. The feed and checksums are signed.
Related
NVIDIA's open stack for running OpenClaw, Hermes and LangChain Deep Agents in OpenShell sandboxes with network policy and managed inference
Static code analysis with rules that look like source code, plus a built-in MCP server for AI agents
Trail of Bits skills
Trail of Bits Skills Marketplace
Trail of Bits plugin marketplace for security work: smart contracts, C/C++ and Rust review, Semgrep, CodeQL and fuzzing
Nine skills that let an agent run Strix pentests, fix validated findings and add security scanning to CI