Security context

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

alirezarezvani/claude-skills

              main @ 4a3c05b
            

Fixes

CVEs

HIGH

Peak severity

Highlights

Command Injection: 1 prior fix. Scrutinize any change in this area.

agents: most-fixed (1 issue). Treat as high-risk during review.

3 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.

Command Injection: The use of shell string execution in subprocess calls allowed arbitrary command injection during automated experiment runs. Transitioning to list-based arguments prevents shell parsing of untrusted inputs.

Sandbox Bypass: Unrestricted capabilities in Claude sub-agent tools allowed execution of dangerous system operations. Restricting capabilities via frontmatter configuration and tool schemas establishes a critical sandbox boundary.

Insecure Deserialization: The use of dynamic evaluation functions like eval and exec on untrusted inputs poses an immediate risk of remote code execution, requiring comprehensive security scanning and replacement with safer parsers.