panopticon-manager is the authorization and dispatch boundary of the
Panopticon pipeline — it decides which typed commands an endpoint agent is
told to execute. Vulnerability reports against this repo (or against the
vendored eyedetect/response_engine submodules as consumed here) are
treated as especially sensitive.
Do not open a public GitHub issue for a security vulnerability. Report privately via GitHub Security Advisories:
https://github.com/Panopticon-Co/panopticon-manager/security/advisories/new
Please include:
- A description of the vulnerability and its impact.
- Reproduction steps, including any sample events, commands, or tokens needed (redacted/synthetic — do not send real credentials).
- Affected file(s)/endpoint(s) and, if known, the specific authorization
invariant being bypassed (see
docs/THREAT_MODEL.mdanddocs/RESPONSE_ENGINE_STATE.md's "Security invariants" section for the invariants this repo currently claims to enforce).
This is a capstone/research project maintained on a best-effort basis — there is no guaranteed response SLA. We will acknowledge reports as soon as reasonably possible and keep the reporter informed as the issue is triaged and, if applicable, fixed.
In scope: panopticon-manager's own code (manager/), and how it consumes
the vendored vendor/eyedetect and vendor/response_engine submodules
(e.g. authorization bypass, command-forgery, cross-agent/cross-analyst
identity confusion, PID-reuse or other target-confusion attacks against
dispatched commands). Vulnerabilities purely internal to the vendored
engines themselves are better reported against their own repositories
(panopticon-detection-engine,
panopticon-response-engine),
though reporting here is also fine if the boundary is unclear.
This project does not yet cut versioned releases; security fixes are applied
to main.