Slipstream
A permission slip before every agent run.
Run a task
The planner proposes a manifest. Nothing runs before you approve it.
Sample dataset · 14 rows · no network
Red team self-test
Canned, harmless scripts that try to get out of the sandbox. Fixed manifest, no model call, and you still approve each run.
Last run audit
Live lifecycle and audit events
IDLE
—Waiting for a task to be planned.
Run report
Task output
Attempts
Final output (stdout of the last attempt)
Audit record (JSON)
How it works
Every task starts as a permission slip that lists the files, memory, CPU, time and websites it may use, and nothing runs until you approve it. The approved work then runs in a locked, throwaway container with a read-only system, no secrets and hard resource limits. It can reach only the websites listed on the slip, through a proxy that refuses everything else. When it finishes, you get a report of what it used and what was blocked, and the container is removed.