Shangjie Chen

refactor: Decompose functions.py into internal modular components (_tool_error_handler, _tool_caller, _batch_tool_executor) and deduplicate single-call async and live tool execution pipelines Co-authored-by: Shangjie Chen <deanchen@google.com> PiperOrigin-RevId: 976053981

fix(callables): extract unified CallableSpec and resolve doc and unwrap edge cases Unify callable introspection, docstring resolution, signature analysis, and type hint caching between FunctionTool and FunctionNode into a shared helper module adk/utils/_callable_utils.py, and place tool argument preprocessing in _schema_utils.py. Key changes: - Add `_callable_utils.py` defining `unwrap_callable`, `get_type_hints_cached`, and `CallableSpec` strictly focused on callable introspection and signature analysis. - Place tool argument preprocessing logic (`preprocess_args`) in `_schema_utils.py`. - Refactor `FunctionTool` to leverage `CallableSpec` for docstring resolution, signature analysis, and type hint caching, and `_schema_utils.preprocess_args` for model coercion. - Refactor `FunctionNode` to use `CallableSpec` for docstring, signature, type hints, and context parameter detection, and cache `spec.unwrapped_func`. - Fix `FunctionTool` description fallback to empty string for undocumented callables whose wrapper docstrings match default object/routine docstrings. - Fix unwrapping loops and signature introspection for builtin functions like `min`, gracefully handling signature-less callables in `FunctionTool.run_async`. - Robust type hint unwrapping: unwrapping `functools.partial` in `CallableSpec` enables resolving annotations under `from __future__ import annotations`, consistently activating Pydantic model coercion in `preprocess_args` where `typing.get_type_hints` previously failed with `TypeError`. - Add comprehensive unit tests in `tests/unittests/utils/test_callable_utils.py` and `tests/unittests/utils/test_schema_utils.py`. Co-authored-by: Shangjie Chen <deanchen@google.com> PiperOrigin-RevId: 975495673

fix: Restore `invocation_context.branch` from session history when executing a non-root agent node or resuming a sub-agent in `Runner` Any event the node authored is eligible, including a plain text turn, which is all a non-resumable sub-agent may leave behind. Tool messages are excluded: a tool's user-facing message is authored under the agent's name and carries the agent's node path, but is published on `<tool>@<function_call_id>`, so a branch whose trailing id matches a function call in the session is skipped. Nodes are matched by static node path (run ids stripped), falling back to author/name for path-less legacy events, disambiguating sub-agents that share a name. At the resume sites the scan is scoped to the invocation being resumed; the new-invocation and live paths use the most recent matching event. This supersedes the earlier (unsubmitted) cl/944151553, whose path-matching and invocation-scoping approach is folded in here. Co-authored-by: Shangjie Chen <deanchen@google.com> PiperOrigin-RevId: 970044047

refactor: Split graph validation and parsing from _graph.py Move graph validation and parsing logic from `_graph.py` to dedicated utility modules (`_graph_validation.py` and `_graph_parser.py`) under `workflow/utils/`. This simplifies the core `Graph` and `Edge` model definitions. Co-authored-by: Shangjie Chen <deanchen@google.com> PiperOrigin-RevId: 941176423

refactor: map YAML agent configs onto agent classes by reflection Each agent class carried its own `_parse_config`, so every config field had to be wired up by hand in the class that consumed it, and a field added to one class was invisible to the others. `_AgentConfigMapper` derives the mapping from the target class instead: it reads the field annotations and resolves each YAML value to whatever that field expects -- callbacks, schemas, tools, sub-agents, models, and `*_code` references. With the mapper deriving all of that, the hand-written parsers on `LlmAgent` and `LoopAgent` have nothing left to add, and go away along with `LlmAgent._resolve_tools`. Existing 1.x configs load unchanged, a subclass overriding `_parse_config` still gets the last word on its kwargs, and the module denylist still gates every code reference. One capability is new: `agent_class` now also names `Workflow` and `FunctionNode`, which head rejected with "Invalid fully qualified name". Reflection maps onto any node class, not just the agent classes that shipped a parser, so restricting it to the latter would have been an extra rule rather than a simpler one. The graph syntax those nodes need is a follow-up. Co-authored-by: Shangjie Chen <deanchen@google.com> PiperOrigin-RevId: 972780349

feat: Implement Workflow as Tool core feature Introduce NodeTool, allowing individual Nodes and entire Workflows to be wrapped and executed as standard ADK Tools. This PR implements the core, single-turn execution capability: - Support auto-wrapping of BaseNode (Workflows) directly in Agent.tools. - Wrapping synchronous and asynchronous function nodes as NodeTools. - Providing a complete sample workflow demonstrating how to run a workflow as a tool. Resumption and multi-turn nested HITL support are skipped in this PR and will be fully enabled in the later PR. Co-authored-by: Shangjie Chen <deanchen@google.com> PiperOrigin-RevId: 943499058

fix: Support multi-turn nested HITL pause and resumption across workflows and agent tools Co-authored-by: Shangjie Chen <deanchen@google.com> PiperOrigin-RevId: 973238301

refactor: extract live flow logic from base_llm_flow into a dedicated module Co-authored-by: Shangjie Chen <deanchen@google.com> PiperOrigin-RevId: 975257909

chore(live): move audio and transcription managers into google.adk.live Migrate `AudioCacheManager`, `AudioCacheConfig`, `TranscriptionManager`, and `AudioTranscriber` from `google.adk.flows.llm_flows` to `google.adk.live`. Backward-compatible re-exports are maintained in `google.adk.flows.llm_flows` for all moved symbols. `RealtimeCacheEntry` also moves, from `google.adk.agents.invocation_context` to `google.adk.live._audio_cache_manager`, because the cache manager is its only real owner. `invocation_context` re-exports it, so that import path keeps working too. This is a pure move: the moved code, including `RealtimeCacheEntry`'s pydantic config, is unchanged, and nothing is added to `google.adk.live.__all__` -- the moved symbols are reached through their module path, so the move commits to no new public API. `google.adk.live` does not import from `google.adk.flows.llm_flows`, so the two modules that needed `require_agent_name` carry a private copy rather than importing upward. Co-authored-by: Shangjie Chen <deanchen@google.com> PiperOrigin-RevId: 972094217

refactor(runners): extract multi-agent routing into agents/_agent_router Extract tree routing algorithms and session branch history recovery (_find_agent_to_run, _is_transferable_across_agent_tree, _can_transfer_between_agents, _restore_branch_from_history) from Runner into a dedicated private helper module src/google/adk/agents/_agent_router.py. Preserve Runner methods as forwarding shims maintaining 100% backwards compatibility. Add unit tests for _agent_router. Co-authored-by: Shangjie Chen <deanchen@google.com> PiperOrigin-RevId: 972918229