Agent mode
The current Agent is a plan-only agent: it produces a structured execution plan but does not perform any actions.
What it can do
- Break a goal into steps
- Suggest usable tools per step (at the planning level)
- Show the tool capability boundary
What it can't do (safety boundary)
- No command execution (no arbitrary shell)
- No server / filesystem access
- No reading of any secret
- No automatic file or database operations
This is intentional: the first Agent is controlled, read-only, and predictable — not an unconstrained executor.
How to use it
- Switch to Agent mode
- Enter a goal
- Review the generated plan steps and suggestions