Security context

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

huggingface/diffusers

              main @ eb0a900
            

Fixes

CVEs

HIGH

Peak severity

60.0%

Coverage

Highlights

Command Injection: 4 prior fixes. Scrutinize any change in this area.

.github/workflows/claude_review.yml: most-fixed (2 issues). Treat as high-risk during review.

8 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.

Remote Code Execution: Attackers can bypass 'trust_remote_code=False' constraints to execute arbitrary Python code when downloading custom models or pipelines from the Hugging Face Hub. This is a recurring architectural flaw in dynamic module loaders.

Command Injection: Evaluating untrusted GitHub context variables (like branch names, file listings, or PR parameters) within 'run' blocks allows contributors to inject malicious shell commands. Using intermediate environment variables is required to mitigate this vector.

Credential Compromise: Improper handling of repository secrets in workflows triggered by external pull requests risks leaking sensitive Hugging Face Hub write and read tokens.