Coding Agent Sandbox Policy
Addition of a second policy system, read in real-time by coding agents via Vibedefend, to specify forbidden patterns (no eval, no child_process.exec, no plain SQL), pre-approved dependencies only, and a 'no plaintext secrets' rule. BEFORE every write operation, the agent (Cursor/Claude Code) receives an allow|deny|warn status for the code action it is currently performing.
Status: Completed
Log in to comment and vote
No comments yet
Be the first to share your thoughts.