Security context

What an agent needs to avoid regressing past fixes and find the next vuln in this repo.

significant-gravitas/autogpt

              master @ 150514f
            

190

Fixes

CVEs

CRITICAL

Peak severity

23.5%

Coverage

Highlights

Auth Bypass: 49 prior fixes. Scrutinize any change in this area.

autogpt_platform/backend/backend/util/request.py: most-fixed (9 issues). Treat as high-risk during review.

102 high-severity fixes in this history; regressions here are high-impact.

Recurring patterns

The bug types that recur here, drawn from past fixes, not open vulnerabilities.

Auth Bypass: Missing ownership validation in graph and node execution paths allows authenticated users to execute unowned graphs or access execution results from other users. This is remediated by validating graph ownership directly against the active user ID.

Server-Side Request Forgery: Attackers continuously bypass SSRF blocklists using DNS rebinding, URL parse confusion, IPv4-mapped IPv6 formats, and custom redirects. Fixes require IP-pinning, IDNA reconstruction validation, and strict redirect-parsing checks.

Command Injection: Insecure subprocess invocations with shell=True and improper parameter validation allow arbitrary shell command execution. Remediations mandate list-based arguments, shlex-parsing, and restriction of commands to isolated Docker environments.