To use ADK tools, users can specify the tool name in a skill object's `additional_tools` and pass the tool in when initializing a SkillToolset.
Co-authored-by: Kathy Wu <wukathy@google.com>
PiperOrigin-RevId: 879230409
This change introduces several improvements to the BigQuery Agent Analytics Plugin:
* **Fix 1 (High):** Error callbacks (`on_model_error_callback`, `on_tool_error_callback`) now emit `status="ERROR"` instead of defaulting to `"OK"`.
* **Fix 2 (Medium):** Schema upgrade now detects missing sub-fields in nested RECORD columns via a new recursive helper. The version label is now stamped only after the `update_table` call succeeds, ensuring failures can be retried.
* **Fix 3 (Medium):** Multi-loop `shutdown()` now drains batch processors on non-current event loops using `run_coroutine_threadsafe` before closing transports.
* **Fix 4 (Medium):** Session state is truncated before logging to prevent oversized payloads.
* **Fix 5 (Low):** String system prompts are now truncated during content parsing.
* **Fix 6 (Low):** Removed the unused `_HITL_TOOL_NAMES` frozenset.
Co-authored-by: Haiyuan Cao <haiyuan@google.com>
PiperOrigin-RevId: 879147684
Link to Issue or Description of Change
1. Link to an existing issue (if applicable):
Closes: #issue_number
Related: #issue_number
2. Or, if no issue exists, describe the change:
Problem:
The agent's specific version wasn't being tracked in our telemetry data, limiting our ability to trace issues to specific agent versions. This change introduces the gen_ai.agent.version attribute to span context, defaulting to an empty string if omitted for backwards compatibility.
Solution:
We want to capture the specific version of an agent during execution by adding an optional version field to the base agent configurations (BaseAgent, BaseAgentConfig).
This solution was chosen because exposing this field directly to OpenTelemetry span attributes (gen_ai.agent.version) ensures the version is automatically recorded alongside other existing metadata (like name and description) during invocation. Defaulting the value to an empty string ensures backwards compatibility without breaking existing agent implementations that do not specify a version.
Testing Plan
- Added test_trace_agent_invocation_with_version to verify that the gen_ai.agent.version attribute is correctly captured when agent.version is populated.
- Updated existing telemetry span tests to ensure gen_ai.agent.version safely defaults to an empty string ('') when no version is provided.
Unit Tests:
- I have added or updated unit tests for my change.
- All unit tests pass locally.
Manual End-to-End (E2E) Tests:
- Tested on Agent Engine and in a local deployment.
Checklist
[x] I have read the CONTRIBUTING.md document.
[x] I have performed a self-review of my own code.
[x] I have commented my code, particularly in hard-to-understand areas.
[x] I have added tests that prove my fix is effective or that my feature works.
[x] New and existing unit tests pass locally with my changes.
[x] I have manually tested my changes end-to-end.
[x] Any dependent changes have been merged and published in downstream modules.
Additional context
Add any other context or screenshots about the feature request here.
Co-authored-by: Achuth Narayan Rajagopal <achuthr@google.com>
PiperOrigin-RevId: 878835568
Merge https://github.com/google/adk-python/pull/4648
**Please ensure you have read the [contribution guide](https://github.com/google/adk-python/blob/main/CONTRIBUTING.md) before creating a pull request.**
### Link to Issue or Description of Change
**1. Link to an existing issue (if applicable):**
- Closes: #4647
- Related: #3429, #3430
**2. Or, if no issue exists, describe the change:**
**Problem:**
`AgentLoader.list_agents()` returns every non-hidden subdirectory in the agents directory, regardless of whether it contains a valid agent definition. This causes non-agent directories (e.g. `tmp/`, `data/`, `utils/`) to appear in the `/list-apps` API response. This affects both the ADK web UI agent selector and any production deployment depending on this API.
**Solution:**
Reuse the existing `_determine_agent_language()` method inside `list_agents()` to verify each candidate directory contains at least one recognized agent file (`root_agent.yaml`, `agent.py`, or `__init__.py`). Directories that fail this check are excluded from the result. This avoids introducing any new methods or abstractions and keeps the check lightweight (filesystem only, no agent imports).
### Testing Plan
**Unit Tests:**
- [x] I have added or updated unit tests for my change.
- [x] All unit tests pass locally.
27 passed in 2.85s:
pytest tests/unittests/cli/utils/test_agent_loader.py -v
======================= 27 passed, 14 warnings in 2.85s ========================
Added `test_list_agents_excludes_non_agent_directories` which creates a temp directory with three valid agent types (package with `__init__.py`, module with `agent.py`, YAML with `root_agent.yaml`) and three non-agent directories, and asserts only the valid agents are listed.
**Screenshots / Video:**
| Before (non-agent directories listed) | After (only valid agents listed) |
|----------------------------------------|----------------------------------|
|<img width="566" height="553" alt="Image" src="https://github.com/user-attachments/assets/0f50084b-319f-480e-8d8a-051c28d4a7e7" />|<img width="567" height="532" alt="Image" src="https://github.com/user-attachments/assets/52d3543f-4c4c-4ff3-a6dd-7d5ce3f19bb2" />|
**Manual End-to-End (E2E) Tests:**
1. Create a project directory containing both valid agent subdirectories and non-agent subdirectories
2. Run `adk web .`
3. Open the web UI and verify only valid agents appear in the agent selector
4. See screenshots below for before/after comparison
### Checklist
- [x] I have read the [CONTRIBUTING.md](https://github.com/google/adk-python/blob/main/CONTRIBUTING.md) document.
- [x] I have performed a self-review of my own code.
- [x] I have commented my code, particularly in hard-to-understand areas.
- [x] I have added tests that prove my fix is effective or that my feature works.
- [x] New and existing unit tests pass locally with my changes.
- [x] I have manually tested my changes end-to-end.
- [ ] Any dependent changes have been merged and published in downstream modules.
### Additional context
COPYBARA_INTEGRATE_REVIEW=https://github.com/google/adk-python/pull/4648 from markadelnawar:fix/list-agents-filter-non-agents-dirs 041895610fa0c52f2bf3cf7ba0d072a5c580c1b6
PiperOrigin-RevId: 878674609
The `_extract_reasoning_value` function now checks for both 'reasoning_content' and 'reasoning' fields in LiteLLM messages, with 'reasoning_content' taking precedence
Close#3694
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 878668213
Merge https://github.com/google/adk-python/pull/4618
## Summary
Fixes#4564
When using `output_key` with a `temp:` prefix (e.g. `output_key='temp:result'`) in a `SequentialAgent`, the output was silently lost. Agent-2 could never read the temp state written by agent-1.
## Root Cause
Two issues in `append_event`:
1. `_trim_temp_delta_state()` removed temp keys from the event delta **before** `_update_session_state()` could apply them to the in-memory session
2. `_update_session_state()` also explicitly skipped `temp:`-prefixed keys
```python
# Before (broken ordering):
async def append_event(self, session, event):
event = self._trim_temp_delta_state(event) # temp keys gone!
self._update_session_state(session, event) # nothing to apply
```
## Fix
Introduce `_apply_temp_state()` which writes temp-scoped keys to the in-memory `session.state` **before** the event delta is trimmed:
```python
# After:
async def append_event(self, session, event):
self._apply_temp_state(session, event) # temp keys → session.state
event = self._trim_temp_delta_state(event) # temp keys removed from delta
self._update_session_state(session, event) # non-temp keys applied
```
This ensures:
- ✅ Temp state is available to subsequent agents within the same invocation
- ✅ Temp state is still stripped from event deltas (not persisted to storage)
- ✅ All three session services (InMemory, Database, SQLite) behave consistently
## Files Changed
- `src/google/adk/sessions/base_session_service.py`: Added `_apply_temp_state()`, reordered `append_event` logic, removed temp-skip in `_update_session_state`
- `src/google/adk/sessions/database_session_service.py`: Added `_apply_temp_state()` call before trim
- `src/google/adk/sessions/sqlite_session_service.py`: Added `_apply_temp_state()` call before trim
- `tests/unittests/sessions/test_session_service.py`: Updated existing test + added new test for sequential agent scenario
## Testing
All 67 session service tests pass across InMemory, Database, and SQLite backends.
COPYBARA_INTEGRATE_REVIEW=https://github.com/google/adk-python/pull/4618 from stakeswky:fix/temp-state-output-key b9fc737e7a6dc07e06e99af3271a8fc026acae4a
PiperOrigin-RevId: 878499263
This change introduces compatibility of the remote_a2a_agent with the new a2a_agent_executor.
New Event Converters:
`to_adk_event.py`= Defines a new set of default converters for transforming A2A types (Task, Message, TaskStatusUpdateEvent, TaskArtifactUpdateEvent) into ADK Event objects.
Configurable Remote Agent:
The A2aRemoteAgentConfig object allows users to override the default event converters with custom ones.
New AgentExecutor Compatibility:
RemoteA2aAgent now checks for `agent_executor_v2` metadata in A2A responses.
If detected, it delegates response handling to a new `_handle_a2a_response_impl` method, which utilizes the modular converters defined in the configuration.
PiperOrigin-RevId: 878487448
Previous rollback CL - cl/872951141
This change introduces a new search_catalog tool within the BigQuery toolset, enabling users to search for BigQuery assets across projects using the Dataplex Catalog API.
Key changes include:
- Adding google-cloud-dataplex as a dependency in pyproject.toml.
- Updating BigQuery credentials to include the Dataplex scope.
- Implementing get_dataplex_catalog_client in client.py to create Dataplex API clients.
- Creating search_tool.py with the search_catalog function, which constructs and executes Dataplex search queries.
- Adding extensive unit tests for the new Dataplex client and the search_catalog tool, covering various scenarios including query filtering and error handling.
- Updating the BigQuery toolset to include the new search_catalog tool.
- Updating the BigQuery samples README to mention the new tool.
PiperOrigin-RevId: 878435463
This change introduces new implementation files for the A2aAgentExecutor and event converters. The existing A2aAgentExecutor now acts as a wrapper, allowing a switch between the legacy and new implementations. The new implementation includes support for execution interceptors and a dedicated executor context.
Main Changes=
`a2a_agent_executor_impl.py` = the new implementation of the AgentExecutor differs from the legacy one (`a2a_agent_executor.py`) for the removal of the TaskResultAggregator and the explicit `InvocationContext` creation. Instead, it uses `ExecutorContext` and delegates event conversion to the new logic that supports streaming. It maintains an `agents_artifact` state map to handle partial updates and emits TaskArtifactUpdateEvents for content. The `long_running_functions.py` is used to keep track of the LongRunning FunctionCalls and respective FunctionResponse, to emit them at the end of the generation loop in a `TaskStateUpdateEvent(input-required/auth-required)`.
`from_adk_event.py` = this file replaces the conversion functions in the `event_converter.py` used to convert the adk events into a2a events, estrapolating them in a dedicated file. The main changes in the methods are the introduction of TaskArtifactUpdateEvent to handle content parts, allowing for true artifact streaming and chunking. It utilizes an `agents_artifacts` dictionary to track artifact IDs across partial events to correctly handle append operations.
PiperOrigin-RevId: 878399140
This change introduces an `ensure_part` helper function that normalizes input to `types.Part`. This allows `save_artifact` methods in `FileArtifactService`, `GcsArtifactService`, and `InMemoryArtifactService` to accept dictionaries, including those with camelCase keys as used by Agentspace, and convert them into proper `types.Part` instances before saving
Close#2886
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 878131948
This change enables round-tripping of EventCompaction data by storing it within the event's custom_metadata under the key "_compaction" when appending events. When retrieving events, the "_compaction" data is extracted from custom_metadata and used to populate the EventActions.compaction field. This is a temporary measure until the Vertex AI SDK's SessionEvent model supports a dedicated compaction field.
Close#3465
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 878128265
Only acquire FOR UPDATE locks on app and user state rows when the event's state_delta contains changes for those specific scopes. This avoids unnecessary locking on state rows that are not being modified, improving concurrency.
Close#4655
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 878108562
details:
* Uses GEPA (https://gepa-ai.github.io/gepa/) to optimize the instructions for the root agent. Can be extended to sub-agents and other components in the future.
* GEPA package is imported dynamically; you do not need to install it along with ADK unless you plan to use this optimizer.
Co-authored-by: Keyur Joshi <keyurj@google.com>
PiperOrigin-RevId: 878009649
Refactor ToolResultBlockParam content handling to use json.dumps for dict/list results.
Implement _generate_content_streaming to handle Anthropic's streaming API
Close#3250
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 877613612
- **Fork-safety (#4636):** Adds PID tracking to `BigQueryAgentAnalyticsPlugin` so that forked child processes detect stale gRPC channels and re-initialize instead of deadlocking. Uses the standard `os.getpid()` pattern (same as SQLAlchemy, gRPC-python).
- **Auto-create views (#4639):** After ensuring the `agent_events` table exists, automatically creates 15 per-event-type BigQuery views (`v_llm_request`, `v_tool_completed`, etc.) that unnest JSON columns into typed, queryable columns. Controlled by `BigQueryLoggerConfig.create_views` (default `True`), idempotent via `CREATE OR REPLACE VIEW`.
- **Trace-ID continuity & o11y alignment (#4645):** Fixes trace_id fracture between early events (USER_MESSAGE_RECEIVED, INVOCATION_STARTING) and later events (AGENT_STARTING onwards) when no ambient OTel span exists. Also aligns BQ rows with Cloud Trace span IDs when o11y is active.
- **Span-ID consistency under ambient OTel (#4640 review):** Fixes `*_STARTING` / `*_COMPLETED` events producing mismatched span IDs when an ambient OTel span is active. Completion callbacks now check for ambient spans and defer to `_resolve_ids` Layer 2 instead of overriding with plugin-synthetic IDs.
- **Stack leak safety (#4640 review):** Adds `TraceManager.clear_stack()` and makes `ensure_invocation_span()` clear stale records from *different* invocations, preventing span stack leaks across invocations. Uses `_active_invocation_id_ctx` to distinguish stale leak vs same-invocation re-entry.
- **Root agent name staleness fix:** `init_trace()` now refreshes `_root_agent_name_ctx` unconditionally on each invocation (previously set-once-on-None). `after_run_callback` resets it alongside other invocation cleanup.
- **Exception-safe cleanup:** `after_run_callback` uses `try/finally` to guarantee invocation state (`clear_stack`, `_active_invocation_id_ctx`, `_root_agent_name_ctx`) is always reset, even if `_log_event` raises.
- **`on_tool_error_callback` span fix:** Previously discarded the span_id from `pop_span()`, causing TOOL_ERROR events to get the wrong span. Now captures and uses the popped span_id.
Co-authored-by: Haiyuan Cao <haiyuan@google.com>
PiperOrigin-RevId: 877580395
This change updates the part_converter to ensure that the name field in a2a_types.FileWithUri and a2a_types.FileWithBytes is correctly mapped to the display_name field in genai_types.FileData and genai_types.Blob, respectively, during conversions between A2A and Genai Part types. Tests are updated to verify this propagation in both directions.
PiperOrigin-RevId: 877507283
LiteLlm provides built-in handling for tool and response format compatibility across different providers, allowing output schemas to be used reliably with tools for any LiteLlm instance
Close#3969
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 877471153
When handling a /run_sse request that includes a functionCallEventId, do not split events that contain both content and artifactDelta
Close#4487
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 877435561
Escape Click’s Wrapping in `adk deploy agent_engine` example
Merge https://github.com/google/adk-python/pull/4337
### Link to Issue or Description of Change
**2. Or, if no issue exists, describe the change:**
**Problem:**
`adk deploy agent_engine` help message loses newlines in docstring formatting
<img width="749" height="257" alt="スクリーンショット 2026-01-31 13 29 18" src="https://github.com/user-attachments/assets/ede7d9e7-609d-4412-acce-c80e24ec1e2f" />
**Solution:**
Add `\b` on a line by itself before the formatted block
### Testing Plan
This is format improvement of help message, so I think there is no need to add test case.
**Unit Tests:**
- [ ] I have added or updated unit tests for my change.
- [x] All unit tests pass locally.
```
% pytest tests/unittests/cli # Python 3.13.8
======================= 260 passed, 140 warnings in 7.73s ========================
```
**Manual End-to-End (E2E) Tests:**
Ran `adk deploy agent_engine --help`, then saw
```
Example:
# With Express Mode API Key
adk deploy agent_engine --api_key=[api_key] my_agent
# With Google Cloud Project and Region
adk deploy agent_engine --project=[project] --region=[region]
--display_name=[app_name] my_agent
```
### Checklist
- [x] I have read the [CONTRIBUTING.md](https://github.com/google/adk-python/blob/main/CONTRIBUTING.md) document.
- [x] I have performed a self-review of my own code.
- [x] I have commented my code, particularly in hard-to-understand areas.
- [x] I have added tests that prove my fix is effective or that my feature works.
- [x] New and existing unit tests pass locally with my changes.
- [x] I have manually tested my changes end-to-end.
- [x] Any dependent changes have been merged and published in downstream modules.
### Additional context
Same solution as #4258
COPYBARA_INTEGRATE_REVIEW=https://github.com/google/adk-python/pull/4337 from ftnext:escape-wrapping-deploy-agent-engine-example 09440380dd5b1e14a48151eba1b808def8ae3b6a
PiperOrigin-RevId: 877205878
Merge https://github.com/google/adk-python/pull/4462
**Please ensure you have read the [contribution guide](https://github.com/google/adk-python/blob/main/CONTRIBUTING.md) before creating a pull request.**
### Link to Issue or Description of Change
**2. Or, if no issue exists, describe the change:**
**Problem:**
I found a typo: arethemselves
>/.../adk-python/src/google/adk/a2a/converters/event_converter.py:245: UserWarning: [EXPERIMENTAL] convert_a2a_message_to_event: ADK Implementation for A2A support (A2aAgentExecutor, RemoteA2aAgent and corresponding supporting components etc.) is in experimental mode and is subjected to breaking changes. A2A protocol and SDK arethemselves not experimental. Once it's stable enough the experimental mode will be removed. Your feedback is welcome.
**Solution:**
Just fix
### Testing Plan
This is typo fix
**Unit Tests:**
N/A
**Manual End-to-End (E2E) Tests:**
N/A
### Checklist
- [x] I have read the [CONTRIBUTING.md](https://github.com/google/adk-python/blob/main/CONTRIBUTING.md) document.
- [x] I have performed a self-review of my own code.
- [ ] I have commented my code, particularly in hard-to-understand areas.
- [ ] I have added tests that prove my fix is effective or that my feature works.
- [ ] New and existing unit tests pass locally with my changes.
- [ ] I have manually tested my changes end-to-end.
- [ ] Any dependent changes have been merged and published in downstream modules.
COPYBARA_INTEGRATE_REVIEW=https://github.com/google/adk-python/pull/4462 from ftnext:fix-typo-a2a-experimental-warning a117314eb524dd93351e17f2183eea080225e43e
PiperOrigin-RevId: 877095788
1. For prototyping and testing purposes, sandbox name can be provided, and it will be used for all requests across the lifecycle of an agent
2. If no sandbox name is provided, agent engine name will be provided, and we will automatically create one sandbox per session, and the sandbox has TTL set for a year.
If the sandbox stored in the session hits the TTL, it will not be in "STATE_RUNNING" so a new sandbox will be created.
Co-authored-by: Lusha Wang <lusha@google.com>
PiperOrigin-RevId: 876450610
details:
* Enables the use of ADK evaluations via LocalEvalService for optimizing agents.
* Provides flexibility in choosing eval sets and eval cases for training and validation.
* Converts ADK eval results into a compact format useful for whitebox agent optimization.
Co-authored-by: Keyur Joshi <keyurj@google.com>
PiperOrigin-RevId: 875818012
Merge https://github.com/google/adk-python/pull/4572
### Link to Issue or Description of Change
**1. Link to an existing issue (if applicable):**
- Closes: #4496
**2. Or, if no issue exists, describe the change:**
**Problem:**
`LoggingPlugin` overrides 12 callback methods from `BasePlugin` but none use the `@override` decorator. Every other plugin in the package (`DebugLoggingPlugin`, `ReplayPlugin`, `RecordingsPlugin`, `EnsureRetryOptionsPlugin`, `_RequestIntercepterPlugin`) already follows this practice. With `mypy --strict` enabled in `pyproject.toml`, missing `@override` means renamed or removed base-class methods would be silently missed in `LoggingPlugin` while being caught everywhere else.
**Solution:**
Import `override` from `typing_extensions` and decorate all 12 overridden callbacks. Purely additive: one import line and 12 decorators. No behavioral, API, or runtime change.
### Testing Plan
**Unit Tests:**
- [ ] I have added or updated unit tests for my change.
- [x] All unit tests pass locally.
No new tests are required — `@override` is a static-analysis-only decorator with no runtime effect. Ran `mypy` on the file before and after the change: same preexisting warnings, no new errors introduced. CI will validate via the existing test suite and linting checks.
**Manual End-to-End (E2E) Tests:**
Not applicable. This change adds only decorators with no runtime behavior. Verified by comparing `mypy` output before and after — no new errors.
### Checklist
- [x] I have read the [CONTRIBUTING.md](https://github.com/google/adk-python/blob/main/CONTRIBUTING.md) document.
- [x] I have performed a self-review of my own code.
- [x] I have commented my code, particularly in hard-to-understand areas.
- [ ] I have added tests that prove my fix is effective or that my feature works.
- [x] New and existing unit tests pass locally with my changes.
- [ ] I have manually tested my changes end-to-end.
- [ ] Any dependent changes have been merged and published in downstream modules.
### Additional context
I am the author of the original issue (#4496). A previous PR (#4544) was opened but is pending clarification, so I'm
submitting this complete PR as the original issue author.
COPYBARA_INTEGRATE_REVIEW=https://github.com/google/adk-python/pull/4572 from cchinchilla-dev:feat/add_override_decorators_to_loggingplugin_4496 142ed872bee39db782c1dccb84f40906ee849bb8
PiperOrigin-RevId: 875811966
This change adds a new `a2a` subpackage with configuration and utility functions for intercepting requests and responses in `RemoteA2aAgent`. The `RemoteA2aAgent` now accepts an `A2aRemoteAgentConfig` to register `RequestInterceptor` instances, allowing custom logic to be executed before and after the A2A message send.
PiperOrigin-RevId: 875559286
invocation_id is no longer required in resuming case, unless no new_message is provided.
Co-authored-by: Shangjie Chen <deanchen@google.com>
PiperOrigin-RevId: 875432024
This change makes DEFAULT_SKILL_SYSTEM_INSTRUCTION raise a UserWarning when accessed, indicating that its content is experimental and subject to change in minor/patch releases. The constant is also made "private" internally.
Co-authored-by: Haiyuan Cao <haiyuan@google.com>
PiperOrigin-RevId: 875288217
Merge https://github.com/google/adk-python/pull/4531
**Problem:**
JSON Schema allows `true` and `false` as valid boolean schemas, where `true` accepts any value and `false` rejects all values. Some MCP servers use this pattern for unconstrained fields. E.g. [mcp-grafana](https://github.com/grafana/mcp-grafana) - see [grafana-mcp-list-tools.json](https://github.com/user-attachments/files/25392430/grafana-mcp-list-tools.json) which was obtained from `tools/list`
The schema sanitizer previously passed booleans through unchanged, causing a Pydantic ValidationError when `_ExtendedJSONSchema` tried to validate them as schema objects.
```
1 validation error for _ExtendedJSONSchema
properties.data.items.properties.model
Input should be a valid dictionary or object to extract fields from [type=model_attributes_type, input_value=True, input_type=bool]
For further information visit https://errors.pydantic.dev/2.12/v/model_attributes_type
Traceback (most recent call last):
...
File "/.foo/.venv/lib/python3.13/site-packages/google/adk/runners.py", line 561, in run_async
async for event in agen:
yield event
File "/.foo/.venv/lib/python3.13/site-packages/google/adk/runners.py", line 549, in _run_with_trace
async for event in agen:
yield event
File "/.foo/.venv/lib/python3.13/site-packages/google/adk/runners.py", line 778, in _exec_with_plugin
async for event in agen:
...<64 lines>...
yield event
File "/.foo/.venv/lib/python3.13/site-packages/google/adk/runners.py", line 538, in execute
async for event in agen:
yield event
File "/.foo/.venv/lib/python3.13/site-packages/google/adk/agents/base_agent.py", line 294, in run_async
async for event in agen:
yield event
File "/.foo/.venv/lib/python3.13/site-packages/google/adk/agents/llm_agent.py", line 468, in _run_async_impl
async for event in agen:
...<5 lines>...
should_pause = True
File "/.foo/.venv/lib/python3.13/site-packages/google/adk/flows/llm_flows/base_llm_flow.py", line 427, in run_async
async for event in agen:
last_event = event
yield event
File "/.foo/.venv/lib/python3.13/site-packages/google/adk/flows/llm_flows/base_llm_flow.py", line 446, in _run_one_step_async
async for event in agen:
yield event
File "/.foo/.venv/lib/python3.13/site-packages/google/adk/flows/llm_flows/base_llm_flow.py", line 578, in _preprocess_async
await tool.process_llm_request(
tool_context=tool_context, llm_request=llm_request
)
File "/.foo/.venv/lib/python3.13/site-packages/google/adk/tools/base_tool.py", line 129, in process_llm_request
llm_request.append_tools([self])
~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^
File "/.foo/.venv/lib/python3.13/site-packages/google/adk/models/llm_request.py", line 255, in append_tools
declaration = tool._get_declaration()
File "/.foo/.venv/lib/python3.13/site-packages/google/adk/tools/mcp_tool/mcp_tool.py", line 200, in _get_declaration
parameters = _to_gemini_schema(input_schema)
File "/.foo/.venv/lib/python3.13/site-packages/google/adk/tools/_gemini_schema_util.py", line 218, in _to_gemini_schema
json_schema=_ExtendedJSONSchema.model_validate(sanitized_schema),
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^^^^^^^^^^^^^^^^^^
File "/.foo/.venv/lib/python3.13/site-packages/pydantic/main.py", line 716, in model_validate
return cls.__pydantic_validator__.validate_python(
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~^
obj,
^^^^
...<5 lines>...
by_name=by_name,
^^^^^^^^^^^^^^^^
)
^
pydantic_core._pydantic_core.ValidationError: 1 validation error for _ExtendedJSONSchema
properties.data.items.properties.model
Input should be a valid dictionary or object to extract fields from [type=model_attributes_type, input_value=True, input_type=bool]
For further information visit https://errors.pydantic.dev/2.12/v/model_attributes_type
```
**Solution:**
Convert boolean schemas to `{"type": "object"}` as the closest approximation available in Gemini's schema model.
Co-authored-by: Xuan Yang <xygoogle@google.com>
COPYBARA_INTEGRATE_REVIEW=https://github.com/google/adk-python/pull/4531 from onematchfox:fix-gemini-schema-bool 383ac0c0c3ab78d77be4503f5d6b9ad26c41b0db
PiperOrigin-RevId: 875219362
When converting an A2A TextPart to a GenAI Part, extract the 'thought' field from the TextPart's metadata and include it in the GenAI Part.
PiperOrigin-RevId: 875129430
Introduces RunSkillScriptTool to execute scripts located in a skill's scripts/ directory.
The execution logic is isolated within a dedicated SkillScriptCodeExecutor wrapper instantiated by RunSkillScriptTool. This wrapper manages script materialization in a temporary directory and executes Python (via runpy) or Shell scripts (returning standard output or JSON-encoded envelopes).
This isolation eliminates the need to modify the underlying `BaseCodeExecutor` interface or implementations (`unsafe_local_code_executor`, etc.) to support working directories or file paths.
Co-authored-by: Haiyuan Cao <haiyuan@google.com>
PiperOrigin-RevId: 875012237
This change introduces a new `client` parameter to the `Gemini` model's constructor. When provided, this preconfigured `google.genai.Client` instance is used for all API calls, offering fine-grained control over authentication, project, and location settings
Close#2560
PiperOrigin-RevId: 874752355
1. For prototyping and testing purposes, sandbox name can be provided, and it will be used for all requests across the lifecycle of an agent
2. If no sandbox name is provided, agent engine name will be provided, and we will automatically create one sandbox per session, and the sandbox has TTL set for a year.
If the sandbox stored in the session hits the TTL, it will not be in "STATE_RUNNING" so a new sandbox will be created.
PiperOrigin-RevId: 874705260
Adds use_id_token and audience fields to ServiceAccount so that
ServiceAccountCredentialExchanger can produce ID tokens instead of
access tokens. This is required for authenticating to Cloud Run, Cloud
Functions, and other Google Cloud services that verify caller identity.
Close#4458
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 874630210
This change introduces a new `client` parameter to the `Gemini` model's constructor. When provided, this preconfigured `google.genai.Client` instance is used for all API calls, offering fine-grained control over authentication, project, and location settings
Close#2560
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 874628604
1. For prototyping and testing purposes, sandbox name can be provided, and it will be used for all requests across the lifecycle of an agent
2. If no sandbox name is provided, agent engine name will be provided, and we will automatically create one sandbox per session, and the sandbox has TTL set for a year.
If the sandbox stored in the session hits the TTL, it will not be in "STATE_RUNNING" so a new sandbox will be created.
Co-authored-by: Lusha Wang <lusha@google.com>
PiperOrigin-RevId: 874415933
Get agents API design is being updated to return full AgentCard instead of agent card url - while it's being rolled out, update get agents method to instantiate agent card from existing fields.
Co-authored-by: Kathy Wu <wukathy@google.com>
PiperOrigin-RevId: 874285065
This change introduces a recursive function to transform JSON schemas to meet OpenAI's strict mode requirements, including adding "additionalProperties: false" to all object schemas, making all properties required, and stripping sibling keywords from $ref nodes. The schema conversion uses deep copies to not mutating the original input
Close#4573
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 874174994
This change allows users to provide a custom PushNotificationConfigStore when converting an ADK agent to an A2A Starlette application. If no custom store is provided, an InMemoryPushNotificationConfigStore is used by default, thios now lets A2A push notification configuration RPCs
Close#4126
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 874118109
The migration from requests to httpx in v1.24.0 broke ApplicationIntegrationToolset because httpx replaces the URL query string when a `params` dict is passed, even if empty. The requests library merged them instead. This extracts any query parameters embedded in the URL path into the explicit params dict before passing to httpx.
Close#4555
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 874112143
This change introduces an interceptor mechanism allowing custom logic to be executed before agent runs, after each event, and after the agent run completes. New dependencies are added to support these features.
PiperOrigin-RevId: 873952199
The default table ID for the BigQueryAgentAnalyticsPlugin is changed from "agent_events_v2" to "agent_events". The class docstring is also updated to remove the "v2.0" reference.
Co-authored-by: Haiyuan Cao <haiyuan@google.com>
PiperOrigin-RevId: 873485931
This CL adds four enhancements to the BigQuery Agent Analytics plugin and fixes a span hierarchy corruption bug.
- **Schema Auto-Upgrade:** Additive-only schema migration that automatically adds missing columns to existing BQ tables on startup. A `adk_schema_version` label on the table (starting at `"1"`, bumped with each schema change) makes the check idempotent — the diff runs at most once per version. Enabled by default (`auto_schema_upgrade=True`) because upgrades are additive-only and fail-safe. Pre-versioning tables (no label) are treated as outdated, diffed, and stamped. No previous schema versions need to be stored; the logic diffs actual columns against the current canonical schema.
- **Tool Provenance:** Adds `tool_origin` to TOOL_* event content, distinguishing six origin types — `LOCAL` (FunctionTool), `MCP` (McpTool), `A2A` (AgentTool wrapping RemoteA2aAgent), `SUB_AGENT` (AgentTool), `TRANSFER_AGENT` (TransferToAgentTool), and `UNKNOWN` (fallback) — via `isinstance()` checks with lazy imports to avoid circular dependencies.
- **HITL Tracing:** Emits dedicated HITL event types (`HITL_CONFIRMATION_REQUEST`, `HITL_CREDENTIAL_REQUEST`, `HITL_INPUT_REQUEST` + `_COMPLETED` variants) for human-in-the-loop interactions. Detection lives in `on_event_callback` (for synthetic `adk_request_*` FunctionCall events emitted by the framework) and `on_user_message_callback` (for `adk_request_*` FunctionResponse completions sent by the user), not in tool callbacks — because `adk_request_*` names are synthetic function calls that bypass `before_tool_callback`/`after_tool_callback` entirely.
- **Span Hierarchy Fix (#4561):** Removes `context.attach()`/`context.detach()` calls from `TraceManager.push_span()`, `attach_current_span()`, and `pop_span()`. The plugin was injecting its spans into the shared OTel context, which corrupted the framework's span hierarchy when an external exporter (e.g. `opentelemetry-instrumentation-vertexai`) was active — causing `call_llm` to be re-parented under `llm_request` and parent spans to show shorter durations than children. The plugin now tracks span_id/parent_span_id via its internal contextvar stack without mutating ambient OTel context.
Co-authored-by: Haiyuan Cao <haiyuan@google.com>
PiperOrigin-RevId: 873114688
Updates _model_response_to_chunk to better handle LiteLLM's streaming delta/message structure, including prioritizing delta when it contains meaningful content and preserving reasoning_content
Close#4225
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 873097502
Client library for the Agent Registry API that allows users to discover, look up, and connect to agents and MCP servers cataloged in the registry.
Co-authored-by: Kathy Wu <wukathy@google.com>
PiperOrigin-RevId: 873073675
In live mode, when the model calls an unregistered tool, ADK now runs on_tool_error_callback before failing. If the callback returns a response, ADK emits
that function response and continues; otherwise it keeps the old ValueError
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 872996178
Previously we resolved tools sequentially by awaiting _convert_tool_union_to_tools() in a loop -- reduce the latency by resolving tools concurrently.
Co-authored-by: Kathy Wu <wukathy@google.com>
PiperOrigin-RevId: 872979105
Merge https://github.com/google/adk-python/pull/4171
**Problem:**
The BigQuery ADK tools currently lack the ability to search for and discover BigQuery assets using the Dataplex Catalog. Users cannot leverage Dataplex's search capabilities within the ADK to find relevant data assets before querying them.
**Solution:**
This PR integrates a new search_catalog_tool into the BigQuery ADK. This tool utilizes the dataplex catalog client library to interact with the Dataplex API, allowing users to search the catalog.
**Unit Tests:**
- [x] I have added or updated unit tests for my change.
- [x] All unit tests pass locally.
Added the screenshots of the manual adk web UI tests - https://docs.google.com/document/d/1c_lMW7NYGKuLAvPFmSkLehbqySeNyXQIhzQlvo3ixmQ/edit?usp=sharing
### Checklist
- [x] I have read the [CONTRIBUTING.md](https://github.com/google/adk-python/blob/main/CONTRIBUTING.md) document.
- [x] I have performed a self-review of my own code.
- [x] I have commented my code, particularly in hard-to-understand areas.
- [x] I have added tests that prove my fix is effective or that my feature works.
- [x] New and existing unit tests pass locally with my changes.
- [x] I have manually tested my changes end-to-end.
- [x] Any dependent changes have been merged and published in downstream modules.
COPYBARA_INTEGRATE_REVIEW=https://github.com/google/adk-python/pull/4171 from sahaajaaa:sahaajaaa-bq-adk 3dbbaa4f909cb25259e8e7d73a00a58fbe9c2f09
PiperOrigin-RevId: 872951141
This change allows `add_memory` to use the `memories.generate` API with `direct_memories_source` when `custom_metadata["enable_consolidation"]` is set to True. This enables server-side consolidation of the provided memories
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 872554004
The `add_memory` methods in `Context` and `BaseMemoryService` now accept `MemoryEntry` objects in addition to strings. The Vertex AI Memory Bank service implementation is updated to handle these new types
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 872108561
* use last-release-sha to locate the previous release commit in main for CHANGELOG generating.
* release tag can now use the commit sha in release branch
Co-authored-by: Wei Sun (Jack) <weisun@google.com>
PiperOrigin-RevId: 871909085
## Problem
When performing authentication flows via `OAUTH2` or `OPEN_ID_CONNECT`, the native `OAuth2Token` response from identity providers, like Google OAuth, often includes an `id_token` alongside the `access_token` and `refresh_token`. [MCP Toolbox](https://googleapis.github.io/genai-toolbox/resources/authservices/google/) implements authentication through ID Tokens and [integrates with ADK](https://google.github.io/adk-docs/integrations/mcp-toolbox-for-databases/) to provide easy tools management for the end-users.
However, the ADK's `update_credential_with_tokens` utility explicitly drops the `id_token`, preventing agents and tools from verifying user identity or extracting OIDC claims securely. Furthermore, the `OAuth2Auth` model does not have a designated field for `id_token`.
## Solution
1. Added an `id_token: Optional[str] = None` field to the `OAuth2Auth` pydantic model in `auth_credential.py`.
2. Updated `update_credential_with_tokens` in `oauth2_credential_util.py` to correctly extract and map `tokens.get("id_token")` into the `OAuth2Auth` credential object.
3. Updated the relevant unit tests to ensure `id_token` is asserted and preserved during credential updates.
### Testing Plan
- I have added or updated unit tests for my change.
- All unit tests pass locally.
PiperOrigin-RevId: 871801313
The list_skills method is not for model tool listing, but for giving the developer flexibility to load the skill name/description at runtime (from discussion in go/orcas-rfc-555)
Co-authored-by: Kathy Wu <wukathy@google.com>
PiperOrigin-RevId: 871406905
This CL fixes several bugs in the BigQuery Agent Analytics plugin and refactors the internal data-passing pattern for better type safety and maintainability.
- **Stale Loop State Validation:** Use `loop.is_closed()` — a public, reliable API — to detect and clean up stale asyncio loop states in `_batch_processor_prop`, `_get_loop_state`, and `flush`. The previous approach used `asyncio.Queue._loop` which is `None` on Python 3.10+, causing the check to always treat states as stale.
- **Quota Project ID Fallback:** Remove the `or project_id` fallback when setting `quota_project_id` on `BigQueryWriteAsyncClient`. This fixes Workload Identity Federation flows where the federated identity lacks `serviceusage.services.use` on the quota project.
- **Kwargs Passthrough:** Pass `**kwargs` through to `_log_event` in all callbacks. Previously only model callbacks forwarded them, causing custom attributes (e.g. `customer_id`) to silently drop for agent, tool, run, and error events.
- **State Delta Logging:** Replace the dead `on_state_change_callback` (never invoked by the framework) with `on_event_callback`, which is already dispatched by the runner for every event. Remove duplicate `STATE_DELTA` logging from `after_tool_callback`.
- **EventData Dataclass:** Replace the `**kwargs`-as-data-bus pattern in `_log_event` with an explicit `EventData` dataclass. This makes the interface self-documenting, catches typos at construction time, and eliminates shared dict mutation across `_resolve_span_ids`, `_extract_latency`, and `_enrich_attributes`. All 12 callback call sites now construct typed `EventData` instances.
- **Multi-Subagent Tool Logging Tests:** Add `TestMultiSubagentToolLogging` (6 tests) verifying that tool events are correctly attributed to subagents in multi-turn, multi-agent scenarios. Total tests: 111 (up from 60).
Co-authored-by: Haiyuan Cao <haiyuan@google.com>
PiperOrigin-RevId: 871381533
When we added the session_lock_map, it resulted in pickling errors during Agent Engine deployment. To fix, we implemented custom getstate and setstate methods to exclude the lock map lock and session lock map from pickling. Closes https://github.com/google/adk-python/issues/4486.
Co-authored-by: Kathy Wu <wukathy@google.com>
PiperOrigin-RevId: 871056554
meanwhile also centralize input-stream creation in registration
Move the LiveRequestQueue stream creation from _call_live
(function_tool.py) to the lazy registration block in
_process_function_live_helper (functions.py). This centralizes the
input_stream: LiveRequestQueue annotation check and stream creation
in one place, and ensures the stream is also recreated on
re-invocation after stop_streaming resets it to None.
_call_live now simply passes the existing .stream if set, without
needing to know about LiveRequestQueue at all.
Co-authored-by: Xiang (Sean) Zhou <seanzhougoogle@google.com>
PiperOrigin-RevId: 869935204
add_events_to_memory now supports memory_write_mode to select generate (event-based extraction/consolidation) or create (direct raw fact writes via memory_facts). This now lets custom memory pipelines while keeping generate as the default path
Co-authored-by: George Weale <gweale@google.com>
PiperOrigin-RevId: 869897256
Using one lock and checking for tables creation instead of schema version.
Closes issue #4445
Co-authored-by: Liang Wu <wuliang@google.com>
PiperOrigin-RevId: 869808097
Merge https://github.com/google/adk-python/pull/4185
**Description**
This PR resolves#4100 by making the `new_message` field optional in the `RunAgentRequest` model.
**The Problem:**
When attempting to resume an agent session via the FastAPI web server, the request would fail with a `422 Unprocessable Entity` if `new_message` was omitted. This prevented "resume-only" workflows where a user just wants to wake up an existing session.
**The Solution:**
Updated `RunAgentRequest.new_message` to be `Optional[types.Content] = None`. The underlying `runner.run_async` logic already supports `None` for resuming purposes, so no further logic changes were required.
**Verification:**
Verified that `RunAgentRequest` now validates successfully when `new_message` is missing, defaulting the field to `None`.
Co-authored-by: Liang Wu <wuliang@google.com>
COPYBARA_INTEGRATE_REVIEW=https://github.com/google/adk-python/pull/4185 from Akshat8510:fix/fastapi-resume-4100 b6d252636aa5f96186507fccf47a278fe733a362
PiperOrigin-RevId: 869761179
previously we only register streaming tool that accept stream input at runner, now uniformly register all streaming tool at runner.
Co-authored-by: Xiang (Sean) Zhou <seanzhougoogle@google.com>
PiperOrigin-RevId: 869447996
currently we started to relay live request to streaming tool even when the tool was not called yet.
Co-authored-by: Xiang (Sean) Zhou <seanzhougoogle@google.com>
PiperOrigin-RevId: 869421826
`_update_type_string` lowercases Gemini-style type strings (e.g.
"STRING" -> "string") for Anthropic API compatibility. It already
recursed into `items` (arrays) and `items.properties` (arrays of
objects), but not into top-level `properties` of object types. This
caused tool schemas with nested object parameters (e.g. Pydantic
models) to produce invalid JSON Schema with uppercase type strings,
resulting in Anthropic API 400 errors.
Recurse into `properties` at every level, which also subsumes the
existing `items.properties` handling.
This can fix unit test github action error by removing `--extra eval`. `--extra a2a` is not needed because it's included in `test`. All tests are still passing.
Co-authored-by: Liang Wu <wuliang@google.com>
PiperOrigin-RevId: 869014318
2026-02-11 21:39:48 -08:00
267 changed files with 31180 additions and 3595 deletions
* Add intra-invocation compaction and token compaction pre-request ([485fcb8](https://github.com/google/adk-python/commit/485fcb84e3ca351f83416c012edcafcec479c1db))
* Use `--memory_service_uri` in ADK CLI run command ([a7b5097](https://github.com/google/adk-python/commit/a7b509763c1732f0363e90952bb4c2672572d542))
* **[Models]**
* Add `/chat/completions` integration to `ApigeeLlm` ([9c4c445](https://github.com/google/adk-python/commit/9c4c44536904f5cf3301a5abb910a5666344a8c5))
* Add `/chat/completions` streaming support to Apigee LLM ([121d277](https://github.com/google/adk-python/commit/121d27741684685c564e484704ae949c5f0807b1))
* Expand LiteLlm supported models and add registry tests ([d5332f4](https://github.com/google/adk-python/commit/d5332f44347f44d60360e14205a2342a0c990d66))
* BigQuery ADK support for search catalog tool ([bef3f11](https://github.com/google/adk-python/commit/bef3f117b4842ce62760328304484cd26a1ec30a))
* Make skill instruction optimizable and can adapt to user tasks ([21be6ad](https://github.com/google/adk-python/commit/21be6adcb86722a585b26f600c45c85e593b4ee0))
* Pass trace context in MCP tool call's `_meta` field with OpenTelemetry propagator ([bcbfeba](https://github.com/google/adk-python/commit/bcbfeba953d46fca731b11542a00103cef374e57))
* **[Evals]**
* Introduce User Personas to the ADK evaluation framework ([6a808c6](https://github.com/google/adk-python/commit/6a808c60b38ad7140ddeb222887c6accc63edce9))
* **[Services]**
* Add generate/create modes for Vertex AI Memory Bank writes ([811e50a](https://github.com/google/adk-python/commit/811e50a0cbb181d502b9837711431ef78fca3f34))
* Add support for memory consolidation via Vertex AI Memory Bank ([4a88804](https://github.com/google/adk-python/commit/4a88804ec7d17fb4031b238c362f27d240df0a13))
* **[A2A]**
* Add interceptor framework to `A2aAgentExecutor` ([87fcd77](https://github.com/google/adk-python/commit/87fcd77caa9672f219c12e5a0e2ff65cbbaaf6f3))
* **[Auth]**
* Add native support for `id_token` in OAuth2 credentials ([33f7d11](https://github.com/google/adk-python/commit/33f7d118b377b60f998c92944d2673679fddbc6e))
* Support ID token exchange in `ServiceAccountCredentialExchanger` ([7be90db](https://github.com/google/adk-python/commit/7be90db24b41f1830e39ca3d7e15bf4dbfa5a304)), closes [#4458](https://github.com/google/adk-python/issues/4458)
* **[Integrations]**
* Agent Registry in ADK ([abaa929](https://github.com/google/adk-python/commit/abaa92944c4cd43d206e2986d405d4ee07d45afe))
* Add schema auto-upgrade, tool provenance, HITL tracing, and span hierarchy fix to BigQuery Agent Analytics plugin ([4260ef0](https://github.com/google/adk-python/commit/4260ef0c7c37ecdfea295fb0e1a933bb0df78bea))
* Change default BigQuery table ID and update docstring ([7557a92](https://github.com/google/adk-python/commit/7557a929398ec2a1f946500d906cef5a4f86b5d1))
* Update Agent Registry to create AgentCard from info in get agents endpoint ([c33d614](https://github.com/google/adk-python/commit/c33d614004a47d1a74951dd13628fd2300aeb9ef))
* **[Web]**
* Enable dependency injection for agent loader in FastAPI app gen ([34da2d5](https://github.com/google/adk-python/commit/34da2d5b26e82f96f1951334fe974a0444843720))
* Add push notification config store to agent_to_a2a ([4ca904f](https://github.com/google/adk-python/commit/4ca904f11113c4faa3e17bb4a9662dca1f936e2e)), closes [#4126](https://github.com/google/adk-python/issues/4126)
* Add support for injecting a custom google.genai.Client into Gemini models ([48105b4](https://github.com/google/adk-python/commit/48105b49c5ab8e4719a66e7219f731b2cd293b00)), closes [#2560](https://github.com/google/adk-python/issues/2560)
* Add support for injecting a custom google.genai.Client into Gemini models ([c615757](https://github.com/google/adk-python/commit/c615757ba12093ba4a2ba19bee3f498fef91584c)), closes [#2560](https://github.com/google/adk-python/issues/2560)
* Check both `input_stream` parameter name and its annotation to decide whether it's a streaming tool that accept input stream ([d56cb41](https://github.com/google/adk-python/commit/d56cb4142c5040b6e7d13beb09123b8a59341384))
* **deps:** Increase pydantic lower version to 2.7.0 ([dbd6420](https://github.com/google/adk-python/commit/dbd64207aebea8c5af19830a9a02d4c05d1d9469))
* edit copybara and BUILD config for new adk/integrations folder (added with Agent Registry) ([37d52b4](https://github.com/google/adk-python/commit/37d52b4caf6738437e62fe804103efe4bde363a1))
* Expand add_memory to accept MemoryEntry ([f27a9cf](https://github.com/google/adk-python/commit/f27a9cfb87caecb8d52967c50637ed5ad541cd07))
* Fix pickling lock errors in McpSessionManager ([4e2d615](https://github.com/google/adk-python/commit/4e2d6159ae3552954aaae295fef3e09118502898))
* fix typo in PlanReActPlanner instruction ([6d53d80](https://github.com/google/adk-python/commit/6d53d800d5f6dc5d4a3a75300e34d5a9b0f006f5))
* handle UnicodeDecodeError when loading skills in ADK ([3fbc27f](https://github.com/google/adk-python/commit/3fbc27fa4ddb58b2b69ee1bea1e3a7b2514bd725))
* Include list of skills in every message and remove list_skills tool from system instruction ([4285f85](https://github.com/google/adk-python/commit/4285f852d54670390b19302ed38306bccc0a7cee))
* Invoke on_tool_error_callback for missing tools in live mode ([e6b601a](https://github.com/google/adk-python/commit/e6b601a2ab71b7e2df0240fd55550dca1eba8397))
* Keep query params embedded in OpenAPI paths when using httpx ([ffbcc0a](https://github.com/google/adk-python/commit/ffbcc0a626deb24fe38eab402b3d6ace484115df)), closes [#4555](https://github.com/google/adk-python/issues/4555)
* Only relay the LiveRequest after tools is invoked ([b53bc55](https://github.com/google/adk-python/commit/b53bc555cceaa11dc53b42c9ca1d650592fb4365))
* Parallelize tool resolution in LlmAgent.canonical_tools() ([7478bda](https://github.com/google/adk-python/commit/7478bdaa9817b0285b4119e8c739d7520373f719))
* race condition in table creation for `DatabaseSessionService` ([fbe9ecc](https://github.com/google/adk-python/commit/fbe9eccd05e628daa67059ba2e6a0d03966b240d))
* Re-export DEFAULT_SKILL_SYSTEM_INSTRUCTION to skills and skill/prompt.py to avoid breaking current users ([40ec134](https://github.com/google/adk-python/commit/40ec1343c2708e1cf0d39cd8b8a96f3729f843de))
* Refactor LiteLLM streaming response parsing for compatibility with LiteLLM 1.81+ ([e8019b1](https://github.com/google/adk-python/commit/e8019b1b1b0b43dcc5fa23075942b31db502ffdd)), closes [#4225](https://github.com/google/adk-python/issues/4225)
* remove duplicate session GET when using API server, unbreak auto_session_create when using API server ([445dc18](https://github.com/google/adk-python/commit/445dc189e915ce5198e822ad7fadd6bb0880a95e))
* Remove experimental decorators from user persona data models ([eccdf6d](https://github.com/google/adk-python/commit/eccdf6d01e70c37a1e5aa47c40d74469580365d2))
* Replace the global DEFAULT_USER_PERSONA_REGISTRY with a function call to get_default_persona_registry ([2703613](https://github.com/google/adk-python/commit/2703613572a38bf4f9e25569be2ee678dc91b5b5))
* **skill:** coloate default skill SI with skilltoolset ([fc1f1db](https://github.com/google/adk-python/commit/fc1f1db00562a79cd6c742cfd00f6267295c29a8))
* Update agent_engine_sandbox_code_executor in ADK ([ee8d956](https://github.com/google/adk-python/commit/ee8d956413473d1bbbb025a470ad882c1487d8b8))
* Update agent_engine_sandbox_code_executor in ADK ([dab80e4](https://github.com/google/adk-python/commit/dab80e4a8f3c5476f731335724bff5df3e6f3650))
* Update sample skills agent to use weather-skill instead of weather_skill ([8f54281](https://github.com/google/adk-python/commit/8f5428150d18ed732b66379c0acb806a9121c3cb))
* update Spanner query tools to async functions ([1dbcecc](https://github.com/google/adk-python/commit/1dbceccf36c28d693b0982b531a99877a3e75169))
* use correct msg_out/msg_err keys for Agent Engine sandbox output ([b1e33a9](https://github.com/google/adk-python/commit/b1e33a90b4ba716d717e0488b84892b8a7f42aac))
* Validate session before streaming instead of eagerly advancing the runner generator ([ab32f33](https://github.com/google/adk-python/commit/ab32f33e7418d452e65cf6f5b6cbfe1371600323))
* **web:** allow session resume without new message ([30b2ed3](https://github.com/google/adk-python/commit/30b2ed3ef8ee6d3633743c0db00533683d3342d8))
### Code Refactoring
* Extract reusable function for building agent transfer instructions ([e1e0d63](https://github.com/google/adk-python/commit/e1e0d6361675e7b9a2c9b2523e3a72e2e5e7ce05))
* add `thinking_config` in `generate_content_config` in example agent ([c6b1c74](https://github.com/google/adk-python/commit/c6b1c74321faf62cc52d2518eb9ea0dcef050cde))
@@ -7,9 +7,9 @@ This sample data science agent uses Agent Engine Code Execution Sandbox to execu
## How to use
* 1. Follow https://cloud.google.com/vertex-ai/generative-ai/docs/agent-engine/code-execution/overview to create a code execution sandbox environment.
* 1. Follow https://docs.cloud.google.com/agent-builder/agent-engine/code-execution/quickstart#create-an-agent-engine-instance to create an agent engine instance. Replace the AGENT_ENGINE_RESOURCE_NAME with the one you just created. A new sandbox environment under this agent engine instance will be created for each session with TTL of 1 year. But sandbox can only main its state for up to 14 days. This is the recommended usage for production environments.
* 2. Replace the SANDBOX_RESOURCE_NAME with the one you just created. If you dont want to create a new sandbox environment directly, the Agent Engine Code Execution Sandbox will create one for you by default using the AGENT_ENGINE_RESOURCE_NAME you specified, however, please ensure to clean up sandboxes after use; otherwise, it will consume quotas.
* 2. For testing or protyping purposes, create a sandbox environment by following this guide: https://docs.cloud.google.com/agent-builder/agent-engine/code-execution/quickstart#create_a_sandbox. Replace the SANDBOX_RESOURCE_NAME with the one you just created. This will be used as the default sandbox environment for all the code executions throughout the lifetime of the agent. As the sandbox is re-used across sessions, all sessions will share the same Python environment and variable values."
@@ -85,11 +85,10 @@ When plotting trends, you should make sure to sort and order the data by the x-a
""",
code_executor=AgentEngineSandboxCodeExecutor(
# Replace with your sandbox resource name if you already have one.
sandbox_resource_name="SANDBOX_RESOURCE_NAME",
# Replace with your sandbox resource name if you already have one. Only use it for testing or prototyping purposes, because this will use the same sandbox for all requests.
This sample demonstrates how to use the `AgentRegistry` client to discover agents and MCP servers registered in Google Cloud.
## Setup
1. Ensure you have Google Cloud credentials configured (e.g., `gcloud auth application-default login`).
2. Set the following environment variables:
```bash
exportGOOGLE_CLOUD_PROJECT=your-project-id
exportGOOGLE_CLOUD_LOCATION=global # or your specific region
```
3. Obtain the full resource names for the agents and MCP servers you want to use. You can do this by running the sample script once to list them:
```bash
python3 agent.py
```
Alternatively, use `gcloud` to list them:
```bash
# For agents
gcloud alpha agent-registry agents list --project=$GOOGLE_CLOUD_PROJECT --location=$GOOGLE_CLOUD_LOCATION
# For MCP servers
gcloud alpha agent-registry mcp-servers list --project=$GOOGLE_CLOUD_PROJECT --location=$GOOGLE_CLOUD_LOCATION
```
4. Replace `AGENT_NAME` and `MCP_SERVER_NAME` in `agent.py` with the last part of the resource names (e.g., if the name is `projects/.../agents/my-agent`, use `my-agent`).
## Running the Sample
Run the sample script to list available agents and MCP servers:
```bash
python3 agent.py
```
## How it Works
The sample uses `AgentRegistry` to:
- List registered agents using `list_agents()`.
- List registered MCP servers using `list_mcp_servers()`.
It also shows (in comments) how to:
- Get a `RemoteA2aAgent` instance using `get_remote_a2a_agent(name)`.
- Get an `McpToolset` instance using `get_mcp_toolset(name)`.
"You have access to tools and sub-agents discovered via Registry."
),
tools=[mcp_toolset],
sub_agents=[remote_agent],
)
Some files were not shown because too many files have changed in this diff
Show More
Reference in New Issue
Block a user
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Luke Street
Kathy Wu
Haiyuan Cao
Google Team Member
Achuth Narayan Rajagopal
Mark Nawar
George Weale
Xiang (Sean) Zhou
不做了睡大觉
Bastien Jacot-Guillarmod
Xuan Yang
Keyur Joshi
Drew Afromsky
Shruti Nair
dependabot[bot]
nikkie
Shangjie Chen
Lusha Wang
Wei (Jack) Sun
Ke Wang
Carlos Chinchilla Corbacho
Wiktoria Walczak
Google Admin
Sasha Sobran
Yifan Wang
Brian Fox
Liang Wu
Sahaja Reddy Pabbathi Reddy
Rohit Yanamadala
Anmol Jaiswal
Akshat8510