Security context

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

langchain-ai/open-swe

              main @ fb467e8
            

Fixes

CVEs

HIGH

Peak severity

56.8%

Coverage

Highlights

Auth Bypass: 17 prior fixes. Scrutinize any change in this area.

agent/reviewer.py: most-fixed (4 issues). Treat as high-risk during review.

24 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: Users could bypass repository-level permissions and view private pull request metadata due to insecure TTL caching or missing request-time permission evaluations. Future changes must avoid caching access rights.

SSRF: Unrestricted HTTP clients and tools used by LLM agents can be exploited to hit internal resources. Insecure redirects, DNS rebinding, and untrusted protocol schemes (like local file/MCP routes) are the main vectors.

Auth Bypass: Untrusted external reviewers on public repositories could run high-privilege agent operations (write-capable runs) or receive highly-scoped GitHub App installation tokens via misconfigured sandbox parameters.