Skip to content

Security: zxhwolfe-dev/aiworkstation-open-source-intelligence

Security

SECURITY.md

Security Policy

Supported surface

Version 0.3 exposes one Skill and nine anonymous data/evidence tools. It has no OAuth, Premium model execution, checkout, entitlement, or payment path. Reports about vulnerabilities in removed future-design code should be checked against the current main tree and release artifact.

Boundaries

  • Tools never clone, install, import, or execute third-party repository code.
  • read_only_hint=true means no user or third-party business data is modified.
  • Requirement search may create, poll, and cancel an ephemeral selector compute task; this is disclosed as a control-plane effect in osi.tool-result.v2.
  • Public source text and license excerpts are untrusted data, not instructions.
  • Missing license evidence never implies permission, and results are not legal advice.
  • Hosted runtime requires an exact release SHA matching the SHA baked into its image.
  • The container binds behind a TLS gateway with request-size, connection, and IP controls.

Provider responses are bounded by size and structure, redirects are restricted, internal fields fail closed, and HTTP responses are explicitly closed. Public errors use osi.error.v2 and do not expose internal exceptions or credentials.

Reporting

Report vulnerabilities privately through the repository security advisory channel. Include the affected version, reproduction, impact, and whether the issue is reachable through local stdio, HTTP, or the Hosted endpoint. Do not include real access credentials or sensitive user data.

Release gates

A candidate is not ready until version/SHA identity, all nine schemas, unit and MCP tests, the built wheel, the Skills ZIP, container configuration, and remote Hosted smoke pass for the same commit. Production must retain the previous image digest for rollback.

There aren't any published security advisories