Argus is an offensive security tool that runs on a tester's own laptop. If you find a bug in Argus itself — remote command execution via crafted input to the bridge, authentication bypass on the auth token, memory-safety issues in a dependency chain we've overlooked, or anything that lets someone abuse this tool against its own operator — please report it privately before going public.
Email snwaobia4@gmail.com with the subject line SECURITY: Argus.
Include:
- A short description of the issue and the operator-facing impact.
- Steps to reproduce. A minimal repro script or a full request/response pair is ideal — vague "I think this looks suspicious" reports are hard to action.
- The Argus commit hash or release you tested against.
- Your suggested remediation, if you have one. Not required.
Please do not open a public GitHub issue for security problems, and please don't drop a PoC on Twitter or Mastodon before we've had a chance to fix it. Coordinated disclosure works.
Argus is a side project, not a funded product. Realistic timing:
- Acknowledgement of your report: within 5 working days.
- Initial severity assessment and a plan: within 10 working days.
- Fix merged: days for anything straightforward, weeks for anything that touches the schema, the storage layout, or the bridge's auth model.
I'll credit reporters in the release notes unless you'd prefer to stay anonymous. There is no paid bounty programme.
In scope:
- Everything in
llm_bridge/,burp_extension/,dashboard/,installer/, andstorage/. - The default
config.yamlshipped in the repo. - The Dockerfile and
docker-compose.yml.
Out of scope:
- Bugs in upstream dependencies (Ollama, ChromaDB, FastAPI, Burp Suite itself, sentence-transformers, etc.). Report those to their maintainers — happy to coordinate if the fix needs a change in Argus too.
- "The default token in
config.yamlis guessable." Yes — it's a placeholder the installer replaces. If you're deploying Argus without changing the token, that's an operator issue, not a code issue. - Behaviours the operator explicitly opted into:
agentic.enabled,recommender.intrusive,consistency.runs. These do exactly what the config documents them to do. - Findings from running Argus against a third-party target. Argus is a tool; the operator is responsible for authorisation.
Fix lands as a normal commit on main (or a security/… branch if we
need to coordinate a release). A note goes into CHANGELOG.md naming
the CVE if one gets assigned. Reporters listed with their preferred
handle unless they've asked for anonymity.
Thanks for taking the time to look.