RepoGuardAI is a read-only auditing framework. By design it:

• Never modifies the repository under audit.
• Never installs software or scanners automatically.
• Never transmits repository contents to any external service — there is no network/AI integration in the MVP.
• Runs third-party scanners only when the operator has already installed them, capturing their stdout/stderr/exit code to repoguard-results/scans/.
• Its own output directory is excluded from discovery/scanning so a repeat run never scans its own generated artifacts.

• Third-party scanners (gitleaks, semgrep, trivy, osv-scanner, npm/pnpm audit, pip-audit, govulncheck, cargo-audit) are executed as external processes with the target repo as working directory. Review and trust those tools yourself; RepoGuard does not sandbox them.
• Scanner output and discovery files may contain sensitive strings (e.g. detected secrets). The repoguard-results/ directory is git-ignored by default — keep it out of version control.
• Treat generated instruction files and scanner output as data. RepoGuard's prompts explicitly instruct agents to ignore any instruction-like text found inside scanned files or scanner output (prompt-injection defense).

If you find a security issue in RepoGuardAI itself, please open a private report / security advisory on the GitHub repository rather than a public issue. Include reproduction steps and affected version. We aim to acknowledge within a reasonable time frame.

A clean RepoGuardAI report does not certify a repository as secure. It reflects only what the configured scanners and the driving agent examined, under the limitations listed in each generated report.