Please do not report security vulnerabilities through public GitHub issues.

Report privately through GitHub's private vulnerability reporting (the Security tab → Report a vulnerability). We aim to acknowledge a report within a few days and will keep you updated as we work on a fix.

When reporting, please include:

• the affected component (backend intel_platform, frontend, or deploy config),
• a description of the issue and its impact,
• steps to reproduce (a proof-of-concept helps), and
• any suggested remediation.

This is an intelligence-analyst workbench that collects untrusted content from the web and renders it back to analysts. Reports touching the trust boundary are especially valuable:

• SSRF / egress — outbound collection fetches are gated through collection/url_guard.py (validate_url) on every request and redirect hop.
• Injection — stored XSS from document- or LLM-derived text, Cypher/SQL injection, prompt injection that escalates privilege.
• AuthN / AuthZ — JWT handling, the admin-gated routes, privilege escalation.
• Secret handling — API keys are Fernet-encrypted at rest; report any leak path.

This project ships with default development credentials and a placeholder JWT_SECRET. Before exposing any instance beyond localhost:

• set REQUIRE_SECURE_AUTH=true — the app then refuses to start while any built-in default secret, API key or admin password is still in place, so the three items below fail loudly rather than silently,
• set strong, non-default admin credentials,
• set a real, high-entropy JWT_SECRET,
• provide real datastore passwords (never the .env.example placeholders), and
• keep .env (and any real keys) out of version control — it is gitignored.

The local docker compose stack binds all services to 127.0.0.1 by design; do not rebind app ports to 0.0.0.0 on an untrusted network.

This is an actively developed project; security fixes land on main. There are no separately maintained release branches — track main for the latest fixes.