If you believe you've found a security issue in this project, do not open a public issue. Use GitHub's private vulnerability reporting:

1. Go to the Security tab of this repository.
2. Click Report a vulnerability.
3. Describe the issue, including reproduction steps and impact assessment.

You'll get an acknowledgement within ~7 days. Realistic timeline for a fix depends on complexity. This is a personal research project, not a 24/7 production service.

This is a personal research codebase. The threat model is narrower than a typical production service:

• In scope: secret leaks in code or commit history, command injection / SSRF / SQL injection in src/, dependency vulnerabilities flagged by gitleaks / Dependabot, RAG-specific issues like prompt injection through indexed corpora that materially compromise the host.
• Out of scope: Denial of service against your own local Docker Compose stack, ML adversarial inputs that produce wrong-but-not-harmful answers, third-party services this project integrates with (report those upstream).

• Secrets never committed to the repo. .env* is gitignored; .env.example is the only template (with placeholder values).
• Pre-commit gitleaks hook + a security GitHub Actions workflow scan every PR for accidental secret commits.
• pyproject.toml author info uses GitHub's no-reply email pattern so contributor identities aren't scraped from package metadata.
• The deploy workflow (.github/workflows/deploy.yml) authenticates to Google Cloud via Workload Identity Federation: short-lived OIDC tokens, no long-lived service-account JSON key stored in GitHub. The WIF provider's attribute-condition scopes the federation to this repo's owner so only its workflows can mint tokens for the deploy service account. Application secrets (for example, OpenRouter API keys for the optional server-side classifier) are injected into Cloud Run via gcloud run services update --update-secrets, sourced from Google Secret Manager, not committed to the repo.