Skip to content

Security: Padraigobrien08/auditable-agent-loop

Security

SECURITY.md

Security Policy

Supported use

This repository is designed to handle:

  • authenticated multi-user API access
  • project/run ownership boundaries
  • persisted artifacts and traceability metadata
  • deployable local or self-hosted environments

Security issues that affect any of those surfaces should be treated seriously.

Reporting a vulnerability

Please do not open a public GitHub issue for a suspected security vulnerability.

Instead, report it privately to the maintainer with:

  • a concise description of the issue
  • the affected surface
  • reproduction steps
  • impact assessment
  • suggested mitigation if you have one

If you already have an established private contact path with the maintainer, use that. If not, open a minimal public issue asking for a private reporting channel without disclosing details.

Good reports include

  • whether the issue affects auth, data isolation, artifact access, secrets, or deployment posture
  • whether it depends on default local-dev configuration
  • whether it affects production-safe settings only, dev settings only, or both
  • exact versions or commit SHA if known

Security priorities for this repo

The highest-priority classes are:

  • cross-project or cross-user data access
  • JWT/session weaknesses
  • unsafe default configuration in deployed environments
  • artifact content disclosure
  • background worker or queue behavior that can corrupt or leak run state
  • prompt/report/trace surfaces exposing secrets or private tenant data

Notes on local development

Some defaults in this repo are intentionally friendlier for local development than for production. That is acceptable only when the boundary is explicit and documented.

A valid security fix in this repo often means:

  • keeping development convenience possible
  • while making production defaults safe and obvious

There aren't any published security advisories