Security context

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

pewdiepie-archdaemon/odysseus

              dev @ 5e9b415
            

173

Fixes

CVEs

CRITICAL

Peak severity

Highlights

Auth Bypass: 62 prior fixes. Scrutinize any change in this area.

core/auth.py: most-fixed (8 issues). Treat as high-risk during review.

99 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: Insecure Direct Object Reference (IDOR) and missing owner-scoping allowed unauthorized users to access deleted document versions or associate sensitive documents with sessions belonging to other tenants. Strict user-ownership checks must be propagated down to every document and session query.

Path Traversal: Agent filesystem tools (like edit_file, grep, glob, and ls) frequently lacked case-insensitive check alignments and strict canonical prefix checks, permitting symbolic links and relative path sequences to traverse outside of the workspace directory.

Indirect Prompt Injection: Untrusted context elements—including email configurations, custom integrations, user-editable skills, and tool results—were directly concatenated into the trusted system prompt instead of being isolated inside untrusted context wrappers.