Human-readable description shown to the LLM so it knows when to use the tool

Execute the tool with its context (for dynamic permission requests) and a string input, returning the result.

Unique name used by the agent to reference this tool

Declared content type of the tool's output (framework carries, does not interpret)

Static resource permissions this tool declares it needs. The runtime requests these through the context before each execution; a denied one short-circuits the call.

Where this tool came from (used by the feedback/adjust system to target patches).

Revert/undo changes the tool has made to external resources.

Schema describing the named parameters this tool accepts in its JSON input object. Empty means the tool takes no parameters (or a single free-form string input).

Verify the output is correct given the input. Returns Ok or Error with reason.

Optional version identifier (e.g. "1.0"). None = unversioned; matches any requested version.

Whether this tool declares revert capability

Whether this tool declares verify capability

Run the tool: request each declared static permission through the context first, then execute. A denied declared permission short-circuits with a refusal message instead of running the tool.

Create a tool that receives its execution context (to request permission dynamically) and declares the static permissions it needs (auto-requested before each run).

Create a simple tool with just name, description, and execute (text/plain, no revert). The execute function ignores the context; use the 4-argument overload for tools that request permission dynamically or declare static permissions.