Tab: Severity policy
Path: AI Foundation > AI Agent > Severity policy
What it shows: how the AI agent treats each tool, based on the severity the tool declares.
| Severity | Behaviour in the AI agent |
|---|---|
| Read | Runs automatically. Results appear as tool-call steps in the chat. |
| Write | Prepares a change card. The change is written through the TYPO3 DataHandler only after you click Apply. |
| Destructive | Red change card with a two-step confirmation (Confirm (1 of 2), then Apply (2 of 2)). |
| Unclassified | Locked. The tool cannot run until its extension declares a severity. |
Below the table, one card per severity lists the tools and the extension that provides them.
What you do here: check which tools can change data before you grant the AI agent to editors. To block individual tools, use the settings described in Tab: Governance.
Important
Write and destructive tools never write on their own. The AI agent always shows a change card and waits for your explicit approval.