Security context
What an agent needs to avoid regressing past fixes and find the next vuln in this repo.
nvidia/nemoclaw
main @ 06db704
86
Fixes
2
CVEs
HIGH
Peak severity
8.0%
Coverage
Highlights
Privilege Escalation: 15 prior fixes. Scrutinize any change in this area.
scripts/nemoclaw-start.sh: most-fixed (9 issues). Treat as high-risk during review.
31 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.
Information Disclosure: Host environment variables and sensitive API tokens are prone to leaking into the untrusted sandboxed agent runtime unless filtered through strict allowlists. CVE-2026-24222 notes remote exfiltration of host environment variables via sandbox initialization.
SSRF: Weaknesses in the endpoint validation component allow attackers to bypass private IP/host restrictions (such as accessing IANA-reserved ranges or 0.0.0.0/8). CVE-2026-24231 tracks this SSRF vector.
Command Injection: Executing shell commands via shell interpolation allows command injection from untrusted arguments (e.g., custom sandbox names or skill paths). Hardening requires strictly transitioning to structured argument arrays.