This is a local, single-user CLI tool. There is no hosted service and no network surface beyond outbound calls to the Google Gemini API.

Please report security issues privately — do not open a public issue for a suspected vulnerability.

• Preferred: open a GitHub private security advisory.

We aim to acknowledge within 5 working days and to provide a remediation timeline within 15 working days. Please allow a 90-day coordinated disclosure embargo before any public discussion, extended only if a fix is demonstrably in progress.

When reporting, include: affected version/commit, reproduction steps, impact, and any suggested mitigation.

This tool requires a GOOGLE_API_KEY (Gemini). Because a single run can fan out to many paid API calls, key hygiene matters:

• Never commit a real key. .env is git-ignored; only .env.example (placeholder) is tracked. Verify with git grep -i AIza before any commit.
• Never log the key. The key is read from the environment and is never printed to stdout/stderr or written to any report. Structured logs redact it.
• Use a project-scoped key, not an account-wide one, so it can be revoked without collateral.
• Rotate at least every 90 days, and immediately if exposure is suspected.
• Set a hard ceiling at the provider. Configure a Google Cloud billing budget / quota so a misconfigured loop cannot run up unbounded cost. The CLI also accepts --max-cost-usd <n> as a client-side soft cap (see README).

Images you pass in are sent to the Google Gemini API for analysis. They leave your machine only via that call.

• De-identify first. You are responsible for stripping PHI / patient identifiers (including EXIF/DICOM metadata) before submission. The tool does not do this for you.
• Compliance is the operator's responsibility. If you feed in real patient data, you — not this project — are responsible for HIPAA (45 CFR Part 164), GDPR Art. 9 (special-category health data), and any local equivalents. Gemini's data-retention terms are governed by your contract with Google.
• This software is not a medical device and must not be used for clinical decision-making. See the disclaimer in README.md.

• Path-traversal protection on all filesystem reads (src/infrastructure/file-scanner.ts).
• Input image extension allow-list (.png, .jpg, .jpeg).
• Text context truncated to 2000 characters to limit prompt-injection surface.
• Images resized to ≤1024px before submission (payload-size bound).
• Mandatory, type-enforced disclaimer on every emitted report.

See docs/SECURITY_CHECKLIST.md for the full audit.