Skip to content

Security: ObtuseAI/blunder

SECURITY.md

Security policy

Reporting a vulnerability

Use this repository's Security tab and Report a vulnerability to open a private GitHub security advisory. Do not open a public issue or pull request for a suspected vulnerability, leaked credential, unsafe authority change, or unintended external effect.

Include the affected revision, component, preconditions, source-to-sink path, impact, and a minimal non-destructive reproducer when practical. Never include live credentials, private memory, raw provider reasoning, browser tokens, or operator approvals.

Good-faith, authorized research that respects the license, avoids privacy violations and service disruption, and gives the maintainers reasonable time to remediate is welcome.

Supported versions

Only the current default branch and latest tagged public preview are supported. Historical private-archive commits, handoff bundles, and generated runtime artifacts are not supported releases.

Security boundaries

  • The core service and desktop sidecar bind to loopback.
  • The React renderer never receives the raw API token and has no arbitrary shell, filesystem, process-control, or direct-network authority.
  • Provider output, retrieved content, reputation, consensus, and fitness are evidence—not permission.
  • Generated validation receives a credential-filtered environment and an audit guard that blocks non-loopback networking, outside-worktree access, child processes, and native-library loading.
  • Windows validation requests a one-process, 768 MiB, kill-on-close Job Object. BLOCKED_BY_PARENT_JOB_POLICY is not described as kernel-enforced isolation.
  • Blunder does not claim AppContainer, virtual-machine, or hostile-native-code containment.

Release hygiene

Before publication, run the native suite, desktop build, dependency audit, current-tree and history secret scans, private-path scan, and exact-SHA GitHub quality workflow. Stop publication and rotate any affected credential if a secret is discovered.

There aren't any published security advisories