ClawSec Scanner
Unify dependency, CVE, source-code, and OpenClaw hook risk scanning in one agent-security workflow.
The evidence shows array-based subprocess arguments, environment-only token handling, report-only defaults, and a signed-release verification procedure. However, the scanner targets an installed skills root by default and can access source, dependencies, and networks; explicit user confirmation, isolation boundaries, sensitive-output handling, rollback, and third-party source verification are incomplete, so points are deducted.
The documentation covers dependencies, warnings, lock handling, and some exception paths. However, SKILL.md describes DAST as static inspection while HOOK.md and the changelog also describe executing real handlers; the handler's timeout option is not enforced by execCommand; and the claimed test files are not included in the supplied evidence. Under static calibration, the score is capped below 10.
The audience, target paths, environment variables, output formats, and OpenClaw event triggers are reasonably specified, with primary fit for OpenClaw. Boundaries for non-OpenClaw platforms, non-POSIX environments, unreachable networks, Chinese-language use, and non-fit cases are weak, and event behavior may cause repeated scans or misuse, so points are deducted.
The skill includes version metadata, a changelog, installation and dependency notes, CLI parameters, troubleshooting, development commands, licensing, and a security-reporting path. Points are deducted because the stated current version and release history are inconsistent, named maintenance ownership and a stable update path are unclear, and static/dynamic-analysis claims conflict.
The design could consolidate npm, Python, OSV/NVD/GitHub, SAST, and OpenClaw hook findings into JSON or text reports. Correctness, completeness, engine availability, and the claimed dynamic-analysis capability cannot be established from static material; missing tools yield partial results and remediation still requires human review, so points are deducted. Static calibration caps the score at 7.
The supplied material includes schemas, source code, a signature-verification procedure, test commands, and changelog entries, providing limited auditability. No actual test suites, CI coverage, or third-party execution results were supplied, and key capability descriptions conflict, so evidence coverage remains thin. Static calibration caps the score at 5.
- Do not treat the HOOK.md claim of real DAST execution as verified; SKILL.md and dast_hook_executor.mjs show static source inspection instead.
- CLAWSEC_SCANNER_TIMEOUT is read and passed but the shown execCommand does not implement timeout termination, so scans may overrun or consume resources indefinitely.
- The default target is the installed skills root and the scanner contacts OSV, NVD, and GitHub; confirm scope, network reachability, token permissions, and possible sensitive-data exposure in reports before deployment.
- The claimed test files and reproducible execution results were not supplied; coverage, false-positive rates, and report correctness require isolated local validation.
What does this skill do, and when should you use it?
ClawSec Scanner is the standalone scanning skill within the broader ClawSec security suite, focused on OpenClaw agent projects. It combines npm audit, pip-audit, OSV, NVD, GitHub Advisory, Semgrep, and Bandit into consolidated vulnerability reports. It also reads OpenClaw HOOK.md metadata and handler source for static hook-risk inspection without executing target code. Scans can run on demand from the CLI or periodically through an OpenClaw event hook.
Scans npm and Python dependencies with npm audit --json and pip-audit -f json; queries OSV and optionally NVD 2.0 and the GitHub Advisory Database; runs Semgrep against JavaScript/TypeScript and Bandit against Python; reads HOOK.md metadata and hook-handler source to identify static risk signals; and emits JSON or text reports containing a scan ID, timestamp, target, vulnerability records, and severity counts. It can also create an OpenClaw hook that scans on agent:bootstrap and command:new events.
- An OpenClaw skill maintainer wants pre-release checks for dependency vulnerabilities and common source-code weaknesses.
- A security engineer needs CVE, GHSA, and SAST findings from several sources normalized into one report.
- An OpenClaw operations team wants scheduled scans with high- and critical-severity summaries posted into conversations.
- A reviewer wants to inspect third-party OpenClaw hooks without importing or executing their handlers.
- A user has an incomplete toolchain but still wants partial coverage, since missing individual tools can be skipped while other scan types continue.
What are this skill's strengths and limitations?
- Combines dependency scanning, CVE enrichment, SAST, and OpenClaw hook inspection in a unified ScanReport.
- Uses OSV as the primary database without authentication, with optional NVD and GitHub Advisory enrichment.
- Usually continues with partial results when the network fails, a single tool is missing, or some JSON is malformed.
- Supports JSON and text output plus event-driven periodic monitoring in OpenClaw.
- The manual release installation includes signed-manifest verification and SHA-256 archive verification.
- Requires a substantial local toolchain, including Node.js, Python, pip-audit, Semgrep, Bandit, jq, and curl.
- Database enrichment requires network access; NVD requests are rate-limited by six seconds without an API key.
- Its DAST capability is static OpenClaw hook inspection, not runtime testing: target handlers are not imported, transpiled, or invoked.
- Missing lockfiles or tools reduce coverage; a nonexistent target, no available scanning tools, or a concurrent scan causes immediate failure.
- The frontmatter reports version 0.0.7 while the roadmap labels v0.0.4 as current, creating version-documentation ambiguity.
How do you install this skill?
Recommended via ClawHub: npx clawhub@latest install clawsec-scanner. The Vercel Skills CLI command is also documented: npx skills add prompt-security/clawsec --skill clawsec-scanner -a openclaw -y. Before installation, verify node, npm, python3, pip-audit, semgrep, bandit, jq, and curl. The documented manual flow downloads a release archive, verifies a signed checksums manifest and SHA-256 archive digest, then installs it. No client-specific installation procedure is documented for other platforms.
How do you use this skill?
For an on-demand scan, set SCANNER_DIR="${INSTALL_ROOT:-$HOME/.openclaw/skills}/clawsec-scanner" and run "$SCANNER_DIR/scripts/runner.sh" --target ./skills/ --output report.json --format json. For a human-readable scan of one directory, use --target ./my-skill/ --format text; to check required tools, use --check. For continuous OpenClaw monitoring, run node "$SCANNER_DIR/scripts/setup_scanner_hook.mjs", restart the OpenClaw gateway, and run /new for an immediate scan. CLAWSEC_SCANNER_INTERVAL controls the interval; CLAWSEC_NVD_API_KEY and GITHUB_TOKEN provide optional API credentials.
How does this skill compare with similar options?
It does not replace npm audit, pip-audit, Semgrep, or Bandit individually; it orchestrates them and normalizes their results while adding OSV/NVD/GitHub Advisory queries and OpenClaw hook inspection. The documentation says traditional web DAST tools such as ZAP and Burp do not apply to these agent platforms, so this skill uses an agent-specific static inspection approach instead.