Agent harness and security
Two lessons on the parts of RobOS that sit between you and a coding agent: a guard that scans prompts, and a daemon that gives an agent a separate home directory.
These are two separate pieces of code. The shared AI library (robos-lib/ai-agent) calls the guard, and the profile daemon is independent of it. Neither is a filesystem or network sandbox. Each lesson says what it checks, where it runs and where it stops.
| Lesson | Code | What it is |
|---|---|---|
| Configure the prompt security guard | packages/robos-lib/prompt-security.js | A regex and entropy scanner that redacts or blocks prompt text before it is sent. |
| Run commands in an ephemeral agent profile | packages/robos-profiled | A daemon that creates a throwaway home directory and environment for a command. |
For which agent CLI answers a request, see AI agents and API mode. For how RobOS picks models for different jobs, see Agent tiers. For the role prompts given to agents, see Choose and customize agent personas.