Security context
What an agent needs to avoid regressing past fixes and find the next vuln in this repo.
tw93/kaku
main @ 11db0cf
26
Fixes
0
CVEs
HIGH
Peak severity
23.8%
Coverage
Highlights
Path Traversal: 6 prior fixes. Scrutinize any change in this area.
kaku-gui/src/ai_tools.rs: most-fixed (5 issues). Treat as high-risk during review.
14 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.
Path Traversal: AI-driven tools (such as fs_read, write, and patch) are prone to directory traversal and sandbox escapes. Failures to strictly validate canonicalized paths against target directory boundaries allow LLM-generated prompts to read and modify arbitrary files on the host filesystem.
Auth Bypass: The manual-approval mechanism gating dangerous AI-generated shell commands can be bypassed using command chaining, subshells, specific evaluation flags (like node or ruby -e), or redirection patterns, leading to unauthorized host mutation.
Information Disclosure: Sensitive credentials, authorization tokens, and CSRF tokens were historically exposed to local processes by passing them as raw command-line arguments to curl, requiring a transition to reading configs from standard input.