ClawSec is an open-source (AGPL) collection of security skills and signed advisory intelligence for AI agent runtimes, from Prompt Security (part of SentinelOne). It protects agents against prompt injection, configuration drift, and malicious or vulnerable skills across four platforms: OpenClaw (and clones such as MoltBot and Clawdbot), NanoClaw, Hermes, and Picoclaw. Each platform has its own entry-point package: clawsec-suite for OpenClaw, clawsec-nanoclaw for NanoClaw, hermes-attestation-guardian for Hermes, and picoclaw-security-guardian for Picoclaw, with Picoclaw self-pen-testing shipped as a separate opt-in package.
ClawSec groups its protections into four layers:
- Signed intelligence: the advisory feed and checksum manifest are verified against a pinned Ed25519 key before published risks are matched against installed skills.
- Guarded installs: an install that matches an advisory stops and needs a second, explicit confirmation before it can continue.
- Integrity and drift: platform-specific baselines for critical agent files, configuration, attestations, and release artifacts, with drift detection against them.
- Audits and reporting: focused audit, posture, self-test, and community-reporting packages where the platform supports them.
ClawSec recommends and gates actions, but destructive removals and install overrides stay under human approval. On OpenClaw, adding the suite and enabling its persistent advisory hook are separate steps, and the hook shows its preflight before it changes any configuration. The *-traffic-guardian packages are specification baselines for platform builders, not shipped runtime proxies.
The security advisory feed consolidates relevant CVEs polled automatically from NIST's National Vulnerability Database, approved community reports, and provisional GitHub advisories that do not yet have CVE identifiers. It is published with a detached signature and the pinned public key, so an agent can verify the feed before acting on it.
