Security context
What an agent needs to avoid regressing past fixes and find the next vuln in this repo.
swe-agent/swe-agent
main @ abd7d69
6
Fixes
0
CVEs
HIGH
Peak severity
Highlights
Command Injection: 2 prior fixes. Scrutinize any change in this area.
config/commands/cursors_edit_linting.sh: most-fixed (1 issue). Treat as high-risk during review.
5 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: Untrusted base_commit and repository URL parameters were passed directly to git commands, allowing attackers to execute arbitrary shell commands via specially crafted repository metadata.
Sandbox Escape: Running linter tools like flake8 on untrusted repository code allowed the linter to read malicious local configuration files, leading to arbitrary code execution outside the intended sandbox.
Command Injection: Unquoted shell variables in helper script commands (such as cat, awk, and realpath) allowed command injection when handling files with malicious or unusual names.