Code Execution
Secure sandbox for running generated code.
Topics
| Document | Description |
|---|---|
| Sandbox | Deno isolation and permissions |
| Worker Bridge | RPC communication between contexts |
| Tracing | Call hierarchy and observability |
Overview
PML can execute agent-generated code safely:
- Sandbox - Deno workers with restricted permissions
- Worker Bridge - RPC protocol for tool injection
- Tracing - Full observability of execution flow
This enables behavioral emergence - agents can write code that combines MCP tools in novel ways.