securityAuditor

import { securityAuditor } from '@agentskit/skills'

const runtime = createRuntime({
  adapter,
  skills: [securityAuditor],
})

await runtime.run('Audit the diff on PR #123.')

#What it covers

• Authn / authz boundaries.
• Injection — SQL / NoSQL / command / LDAP / template / prompt injection in LLM-fed content.
• Secrets in code, logs, client bundles, error messages.
• Crypto — weak hashing, password compare timing leaks, predictable randomness.
• Network — SSRF on user-supplied URLs, XXE, missing TLS verification.
• Supply chain — postinstall scripts, typosquats, lockfile drift.
• LLM-specific — untrusted instructions in tool output / RAG context, sandbox escapes, jailbreak surfaces.

#Output

Severity ladder: critical · high · medium · low · info. Every finding cites a real file:line, names the exploit, and proposes a concrete fix.

#vs. prReviewer and codeReviewer

• prReviewer enforces project-Manifesto rules.
• codeReviewer is general-purpose code quality.
• securityAuditor is only security. Use it alongside the others, not instead of them.

#Related

• Skills overview · prReviewer · codeReviewer
• Production → Security