Update experimental JSON schema per issue #1048

[zyantw]

Adds documentation for the experimental JSON schema-based function declaration feature to function-tools.md.

Includes:

• An explanation of the ADK_ENABLE_JSON_SCHEMA_FOR_FUNC_DECL environment variable.
• CLI instructions for enabling and disabling the feature across macOS, Linux, and Windows (CMD & PowerShell) using tabbed code blocks.
• A Python code snippet demonstrates how to explicitly configure the feature in-code.

---

1. JSON Schema Generation Capability & Default Status

Quoted Claim

"ADK automatically generates JSON schemas from your function signatures using Pydantic for model function declarations. This feature (JSON_SCHEMA_FOR_FUNC_DECL) is enabled by default in ADK."

1. FeatureName.JSON_SCHEMA_FOR_FUNC_DECL is registered in _FEATURE_REGISTRY with FeatureConfig(FeatureStage.EXPERIMENTAL, default_on=True).
2. When constructing a FunctionTool, _get_declaration() passes is_feature_enabled(FeatureName.JSON_SCHEMA_FOR_FUNC_DECL) to build_function_declaration(), which delegates to _function_tool_declarations.build_function_declaration_with_json_schema().
3. _build_parameters_json_schema() extracts parameter types from the callable using Python inspect and get_type_hints, dynamically generates a Pydantic model via pydantic.create_model, and exports the schema using model_json_schema().

• Citations:
  • src/google/adk/features/_feature_registry.py#L52
  • src/google/adk/features/_feature_registry.py#L174-L176
  • src/google/adk/tools/_automatic_function_calling_util.py#L220-L237
  • src/google/adk/tools/_function_tool_declarations.py#L15-L23
  • src/google/adk/tools/_function_tool_declarations.py#L191-L215
  • src/google/adk/tools/_function_tool_declarations.py#L282-L356
  • src/google/adk/tools/function_tool.py#L145-L156

---

2. Feature Resolution Hierarchy & Environment Variables

Quoted Claim

"You can explicitly control or override this behavior using environment variables or directly in Python code."

• Status: PASS
• Verification Details:
  is_feature_enabled() implements a strict 3-tier priority resolution:
  1. Programmatic Overrides: Checked in _FEATURE_OVERRIDES (set via override_feature_enabled()).
  2. Environment Variables: ADK_ENABLE_<NAME> returns True if set to 1 or true. If not enabled, ADK_DISABLE_<NAME> returns False if set to 1 or true.
  3. Registry Default: Returns config.default_on (True for JSON_SCHEMA_FOR_FUNC_DECL).
• Citations:
  • src/google/adk/features/_feature_registry.py#L258-L284
  • src/google/adk/features/_feature_registry.py#L286-L344
  • src/google/adk/utils/env_utils.py#L27-L60
  • docs/guides/features/feature_registry/index.md#L36-L90

---

3. Option 1: Command Line Configuration (Enable & Disable)

• Verification Details:
  • ADK_ENABLE_JSON_SCHEMA_FOR_FUNC_DECL is formed dynamically via f"ADK_ENABLE_{feature_name_str}".
  • ADK_DISABLE_JSON_SCHEMA_FOR_FUNC_DECL is formed dynamically via f"ADK_DISABLE_{feature_name_str}".
  • is_env_enabled() executes os.environ.get(env_var_name, default).lower() in ['true', '1']. Setting "1" in Bash, CMD, and PowerShell satisfies this requirement.
• Citations:
  • src/google/adk/features/_feature_registry.py#L331-L339
  • src/google/adk/utils/env_utils.py#L27-L60
  • docs/guides/features/feature_registry/index.md#L39-L54

---

4. Option 2 - Method A: Programmatic Feature Override

• Verification Details:
  1. FeatureName and override_feature_enabled are exported in src/google/adk/features/__init__.py.
  2. override_feature_enabled(FeatureName.JSON_SCHEMA_FOR_FUNC_DECL, True) registers the boolean in _FEATURE_OVERRIDES, superseding environment variables and defaults.
  3. FunctionTool(func=...) receives callable get_current_weather, parses its docstring, annotations, and parameters cleanly.
  4. Agent(model="gemini-2.5-flash", name="my_agent", instruction="...", tools=[...]) matches:
     • name: str validated identifier from BaseNode (src/google/adk/workflow/_base_node.py#L48-L56).
     • model: Union[str, BaseLlm] from LlmAgent (src/google/adk/agents/llm_agent.py#L291).
     • instruction: Union[str, InstructionProvider] from LlmAgent (src/google/adk/agents/llm_agent.py#L314).
     • tools: list[ToolUnion] from LlmAgent (src/google/adk/agents/llm_agent.py#L392).
• Citations:
  • src/google/adk/features/__init__.py#L1-L26
  • src/google/adk/features/_feature_registry.py#L258-L279
  • src/google/adk/agents/__init__.py#L29-L62
  • src/google/adk/agents/llm_agent.py#L260-L394
  • src/google/adk/agents/llm_agent.py#L1451
  • src/google/adk/tools/__init__.py#L34-L72
  • src/google/adk/tools/function_tool.py#L99-L143
  • src/google/adk/workflow/_base_node.py#L48-L56

---

5. Option 2 - Method B: In-Script Environment Variable Setting

• Verification Details:
  is_feature_enabled() does not cache initial os.environ lookups at import time; it evaluates os.environ on every invocation when resolving flags. Setting os.environ before tool declaration instantiation takes effect reliably.
• Citations:
  • src/google/adk/features/_feature_registry.py#L331-L336
  • src/google/adk/utils/env_utils.py#L27-L60
  • docs/guides/features/feature_registry/index.md#L99-L100