Agent Security: Prompt Injection, Permissions, Sandboxes, and Credentials
Untrusted content will influence the model, so the execution system must limit what that influence can achieve: what prompt injection, permissions and approvals, sandboxes, credentials, and MCP authorization each cover.
Architecture, Safety, and Basic Commands
Separate local and remote responsibilities, understand connection layers, and verify the correct server before changing it.
Running Subprocesses Safely
Python subprocess invocation, output, timeouts, environments, shell boundaries, and process lifecycle.
SSH Key Access and Hardening
Distinguish user and host keys, then disable password and root login with a multi-session safety procedure.
UFW Without SSH Lockout
Stage temporary and interface-specific rules, verify fresh connections, and then remove public SSH safely.