Sandbox Untrusted Code Execution

About this workflow

Contain what your agent runs: isolated execution environments with resource limits, network policy, and workspace mounting done right.

What happens when you run it

1. 1
   Decide what the boundary must hold before building it.

   Runs flow-sandbox-untrusted-code-execution-1-choose-the-isolation-and-define-the-poli on Claude Sonnet 4.6
2. 2
   Same tool interface, contained blast radius.

   Runs flow-sandbox-untrusted-code-execution-2-build-the-sandboxed-executor on Claude Sonnet 4.6
3. 3
   Default-deny, with every exception named.

   Runs flow-sandbox-untrusted-code-execution-3-enforce-and-verify-the-network-policy on Claude Sonnet 4.6
4. 4
   Trust the boundary only after it has survived you.

   Runs flow-sandbox-untrusted-code-execution-4-attack-it-then-tune-the-ergonomics on Claude Sonnet 4.6

Best for

• •Recurring tasks you currently copy-paste between tools.
• •When a single prompt isn't enough — you need a chain where output(n) feeds prompt(n+1).
• •Fork and adapt if your inputs differ materially from the author's use case.

Models used