FPF Agent Stack

Feature: Tool call validation The executor must treat model output as untrusted input and validate all tool call arguments. @tool @schema Scenario: Unknown tool name is rejected Given FunctionGemma outputs a tool call "delete_host_files" with any arguments When the executor validates the tool call against the loaded skills Then the guard decision should be "abstain" And no tool should be executed @tool @schema Scenario: Invalid arguments are rejected Given the loaded tool "write_file" requires: | field | type | | path | string | | text | string | And FunctionGemma outputs a tool call "write_file" with arguments: | path | 123 | | text | "hi" | When the executor validates the tool call Then the guard decision should be "abstain" And the explanation should mention "schema validation"