Security context
What an agent needs to avoid regressing past fixes and find the next vuln in this repo.
a5c-ai/babysitter
main @ b10d119
3
Fixes
0
CVEs
HIGH
Peak severity
Highlights
Auth Bypass: 1 prior fix. Scrutinize any change in this area.
packages/sdk/src/cli/commands/sessionCreate.ts: most-fixed (1 issue). Treat as high-risk during review.
2 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: Insecure construction of shell arguments executed within bash sessions allowed arbitrary command injection. Developers must use strict shell escaping utilities like shellQuoteArg on all dynamic inputs before execution.
Sandbox Escape: Unrestricted tool availability within commands allowed agent execution to step outside intended sandbox boundaries. Enforcement of explicit tool allowlists in frontmatter configurations is required to mitigate this risk.
Auth Bypass: Prompt injection vulnerabilities allowed LLM agents to automatically bypass interactive approval breakpoints when encountering empty user responses. Robust system prompt constraints are necessary to prevent auto-approval of sensitive tasks.