Skip to content

Security: axlezhao/AlphaLens

SECURITY.md

Security policy

AlphaLens is an experimental beta. It has not undergone an independent security audit and should not be exposed as a public, multi-tenant financial service without additional hardening. Only the latest main and the current beta are considered for fixes; older snapshots have no maintenance guarantee.

Reporting a vulnerability

Use GitHub private vulnerability reporting. Do not open a public issue containing exploit details, credentials or private research. If private reporting is unavailable, use a private contact method listed on the maintainer's profile; do not post sensitive details while waiting for a channel.

Include the affected commit/version, impacted component, a minimal sanitized reproduction, expected/observed behavior and potential impact. Test only against your own isolated environment. Do not probe the hosted preview, access another user's records, or send test payloads to third-party providers without authorization.

There is no bug bounty or guaranteed response SLA. Coordinate disclosure with the maintainer and avoid exposing affected users before a fix or mitigation is available.

Deployment trust boundaries

  • Current session authentication relies on identity headers supplied by a trusted hosting proxy. A self-hosted deployment must strip user-supplied identity headers and validate authentication before supplying trusted identity. Never expose that header trust directly to the internet.
  • Keep provider keys and connector encryption/signing secrets server-side, outside version control. Rotate leaked credentials immediately; deleting a commit is not sufficient.
  • Treat webhook/IR destinations and external documents as untrusted. Current URL checks are not a complete DNS/rebinding/IPv6 defense; use network egress restrictions and harden validation before accepting arbitrary destinations.
  • Skill permission declarations are not a sandbox. Do not install or run untrusted skills with secrets or privileged tools.
  • Multi-tenant access, deletion, worker recovery and webhook delivery require integration/adversarial testing before production use. Local throttles do not enforce a global provider budget.
  • Do not use historical as_of labels as proof of point-in-time correctness, or heuristic quality scores as investment confidence.

See the roadmap and operations guide for known limitations. Code licensing does not grant rights to third-party data.

There aren't any published security advisories