How it works
- The agent submits a script. When no action in its menu fits the job, the agent proposes a script for a host.
- You review it. The script is held until someone reviews it. You read the exact text that will run.
- Approval pins the hash. Approving the script pins its hash. If the text changes, the hash no longer matches and the connector does not run it.
- The connector checks before every run. The connector re-checks the hash before each run, with a 60-second cap per run.
Why not just give it a shell?
A shell lets the agent run anything its account can. A reviewed script lets it run one piece of text you have read, on the hosts its policy reaches, and nothing else.
The record
The submission, the approval and every run go into the keyed, hash-chained audit log, which you can verify offline with your key.
From script to action
A script you approve often enough is a tool waiting to happen. Turn it into a named action with typed arguments, and the agent can call it without a review each time. See CLIs and scripts.