Security skill for coding agents
A security skill for coding agents that answers with the standard, not with an opinion. It turns 29 official standards (OWASP, NIST, MITRE, CWE, CVSS 4.0, SLSA, CISA) into checkable controls, and every result cites the standard, version, identifier and file that support it.
Three modes
| Mode | You ask | You get |
|---|---|---|
| Consultation | "Which controls apply to this API?" | Controls with a state (pass, fail, not applicable, not verified), the evidence and the sources |
| Code audit | "Audit this repository" | Confirmed findings, leads to validate and coverage, starting with a sweep for exposed secrets, keys and published ports |
| Dependency audit | "Anything malicious in our packages?" | An SBOM, known vulnerabilities (OSV, CISA KEV, EPSS), known-malicious packages and typosquats |
The audits end in a validated findings.json and a self-contained HTML report, in English or Spanish, with a copyable mitigation prompt and the critical findings preselected. Secret values are always masked.
Install
In Claude Code it installs as a plugin, since the repository is its own marketplace. Any other agent that reads ~/.claude/skills or ~/.agents/skills can use it through a symlink script, so a git pull keeps it current.