cso

Invoke the cso skill with a phrase like "security audit" or "OWASP review". It detects your stack, then runs phased checks for secrets, dependencies, CI/CD, webhooks, and LLM risks, reporting only findings that pass a confidence threshold.

The secrets archaeology phase searches git history for known credential prefixes like AKIA, ghp_, sk-, and xoxb-, checks whether .env files are tracked, and flags CI configs with inline secrets. Placeholders and test fixtures are excluded as false positives.

Daily mode reports only findings at 8/10 confidence or higher to eliminate noise during regular development. Comprehensive mode is a monthly deep scan with a 2/10 reporting bar that surfaces lower-confidence issues for manual review.

Yes, a dedicated phase checks for user input flowing into system prompts, unsanitized LLM output rendered as HTML, eval of AI responses, unvalidated tool calls, and exposed AI API keys. User content in normal message position is not flagged.

Yes, the skill supply chain phase scans repo-local skills automatically for credential access, network exfiltration, and prompt injection patterns. Scanning globally installed skills requires explicit user permission since it reads files outside the repo.

Package manager audit tools are optional; if one is not installed, the report notes it as skipped with install instructions rather than failing. The audit continues with whichever tools are available and applies false-positive rules to filter noise.