Core tools
The tools an agent can call, and their boundaries.
The agent observes and changes the world through a small set of tools. Each one is authorized independently.
Five base tools
| Tool | Purpose | Risk |
|---|---|---|
search |
Find files by name or content | Low |
read |
Read file contents | Low |
write |
Create or overwrite a file | Medium |
exec |
Run a command | High |
list |
List a directory tree | Low |
Symbol index
The symbols tool reads a code symbol index to locate functions, types and variables quickly. It is based on language parsing rather than text matching, so it beats grep.
Subagents
spawn_agent lets the main agent hand a subtask to a fresh agent instance to work in parallel, then collect the result. See Subagents.
Boundaries
Whatever tool it calls, paths never cross the workspace root. Command execution is likewise bound by permission rules — more tools does not mean more power.