Security context

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

arc53/DocsGPT

              main @ e0aff39
            

Fixes

CVEs

CRITICAL

Peak severity

36.4%

Coverage

Highlights

Information Disclosure: 7 prior fixes. Scrutinize any change in this area.

application/agents/tool_executor.py: most-fixed (3 issues). Treat as high-risk during review.

32 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.

Broken Access Control / Authorization Bypass: Multiple fixes address missing owner-scoped checks or fail-open authorization, allowing cross-tenant or cross-user access to artifacts, conversations, and documents. The breadth of components indicates a systemic issue with enforcing ownership at every data-retrieval and mutation point.

SSRF: SSRF is a recurring high-severity issue across multiple network-facing tools. Fixes centralize URL validation (pinned_request, safe_url) but the existence of separate fixes in api_tool, read_webpage, and mcp_tool suggests that each tool historically had its own ad-hoc checks, creating risk of incomplete enforcement on sibling paths.

Information Disclosure: Information disclosure stems from leaking sensitive data in error messages, logs, kernel environments, and hardcoded credentials. These are distinct sinks but share a theme of failing to redact or scrub secrets from outputs and execution contexts.