Found docs updates needed from ADK python release v2.2.0 to v2.4.0

[xuanyang15]

Compare URL: google/adk-python@v2.2.0...v2.4.0

Feature changes

1. New ManagedAgent feature and documentation

Doc file: docs/agents/managed-agent.md, mkdocs.yml

Current state:

File does not exist. (For docs/agents/managed-agent.md)
Missing from mkdocs.yml navigation.

Proposed Change:

Create docs/agents/managed-agent.md detailing ManagedAgent requirements (enterprise client, global location) and limitations (server-side tools only, streaming only). Add agents/managed-agent.md to mkdocs.yml under Agents -> Simple agents.

Reasoning:
A new ManagedAgent class has been added to ADK for running agents via the Managed Agents API (interactions.create). It requires specific configurations and documentation for users to discover and use it.

Reference: src/google/adk/agents/_managed_agent.py

2. RunConfig field additions (model_input_context, translation_config, http_options, telemetry)

Doc file: docs/runtime/runconfig.md

Current state:

Does not mention model_input_context, translation_config, http_options, or telemetry.

Proposed Change:

Add definitions for:

• model_input_context: Transient context to include in the model input for this invocation without persisting it to the session.
• translation_config: Configures real-time speech-to-speech translation.
• http_options: HTTP options for the agent execution (e.g., custom headers).
• telemetry: Per-request OpenTelemetry configuration.

Reasoning:
New fields were added to RunConfig to support transient model input context, real-time speech translation, custom HTTP options, and per-request telemetry control.

Reference: src/google/adk/agents/run_config.py

3. Support for partial turn updates in streaming requests

Doc file: docs/streaming/dev-guide/part2.md

Current state:

LiveRequest and send_content() documentation lack mention of partial turns.

Proposed Change:

Update LiveRequest to include partial: bool = False # Signal partial turn update. Update send_content() to explain: "You can optionally pass partial=True to indicate that the content is a partial turn update that does not yet complete the model's turn."

Reasoning:
The LiveRequest model and send_content() method were updated to accept a partial boolean flag indicating a partial turn update, which must be documented for developers using the streaming API.

Reference: src/google/adk/agents/live_request_queue.py

4. Deprecation of global_instruction in favor of GlobalInstructionPlugin

Doc file: docs/agents/llm-agents.md

Current state:

Note: For instructions that apply to all agents in a system, consider using global_instruction on the root agent.

Proposed Change:

Note: For instructions that apply to all agents in a system, consider using the GlobalInstructionPlugin on your App object.

Reasoning:
The global_instruction field on LlmAgent has been deprecated. Users should now use the GlobalInstructionPlugin at the App level to apply instructions consistently across all agents.

Reference: src/google/adk/plugins/global_instruction_plugin.py

5. Update ContextCacheConfig parameter documentation

Doc file: docs/context/caching.md

Current state:

• min_tokens (int): The minimum number of tokens required in a request to enable caching... Defaults to 0.

Proposed Change:

• min_tokens (int): The minimum prior-request tokens required to enable caching. This gates on the previous request's actual prompt token count... Gemini enforces a hard 4096-token minimum...
• create_http_options (Python only, google.genai.types.HttpOptions): Optional HTTP options to pass to the GenAI client...

Reasoning:
The ContextCacheConfig definition has been updated to clarify how min_tokens works with the Gemini API and to introduce the create_http_options parameter.

Reference: src/google/adk/agents/context_cache_config.py

6. Document support for Workflows/Nodes as Tools

Doc file: docs/agents/llm-agents.md

Current state:

Lists native functions, BaseTool, and AgentTool as supported tools.

Proposed Change:

Add: * (Python only) An instance of a BaseNode or Workflow that has a description configured (wrapped internally as a NodeTool).

Reasoning:
LlmAgent now supports using BaseNode and Workflow instances directly as tools, expanding ways agents can delegate to other execution graphs.

Reference: src/google/adk/agents/llm_agent.py

7. Deprecate SequentialAgent in favor of Workflow

Doc file: docs/agents/workflow-agents/sequential-agents.md

Current state:

!!! note "Alternative: graph-based workflows"
Starting in ADK 2.0 for Python and Go, templated workflows have been superseded...

Proposed Change:

!!! warning "Deprecation Notice"
SequentialAgent is deprecated in favor of Workflow and will be removed in a future version. Note that Workflow cannot yet be used as an LlmAgent sub-agent.

Reasoning:
SequentialAgent is now explicitly marked as deprecated in Python code. A clear warning should be provided to users.

Reference: src/google/adk/agents/sequential_agent.py

8. Document prompt parameter in OAuth2Auth

Doc file: docs/tools-custom/authentication.md

Current state:

client_secret=YOUR_OAUTH_CLIENT_SECRET

Proposed Change:

client_secret=YOUR_OAUTH_CLIENT_SECRET,

Optional: prompt="consent" (default), "login", "none", or "select_account"

Reasoning:
The OAuth2Auth model now accepts an optional prompt parameter for customizing OAuth prompt behavior.

Reference: src/google/adk/auth/auth_credential.py

9. Add Express Mode API Key support for CLI deploy

Doc file: docs/deploy/agent-runtime/deploy.md

Current state:

adk deploy agent_engine
--project=$PROJECT_ID
--region=$LOCATION_ID
--display_name="My First Agent"
multi_tool_agent

Proposed Change:

With Google Cloud Project and Region

...

Or with Express Mode API Key (bypasses Google Cloud Project/Location requirement)

adk deploy agent_engine
--api_key="YOUR_API_KEY"
multi_tool_agent

Reasoning:
adk deploy agent_engine now supports an --api_key option to automatically deploy using Express Mode, bypassing the need for a GCP project.

Reference: src/google/adk/cli/cli_deploy.py

10. Document trigger sources for ambient agents on Agent Platform

Doc file: docs/runtime/ambient-agents.md

Current state:

Mentions --trigger_sources="pubsub,eventarc" for adk deploy cloud_run.

Proposed Change:

Add note: "You can also deploy ambient agents to Agent Platform by passing the same --trigger_sources flag to the adk deploy agent_engine command."

Reasoning:
adk deploy agent_engine now explicitly supports the --trigger_sources flag for deploying ambient agents.

Reference: src/google/adk/cli/cli_deploy.py

11. Custom Storage Backends via services.yaml

Doc file: docs/sessions/session/index.md

Current state:

Here’s a simplified flow of how Session and SessionService work together

Proposed Change:

Add a section detailing how to register custom storage backends by creating a services.yaml file and specifying the custom URI scheme (e.g., adk web --session_service_uri="mycustom://...").

Reasoning:
The Service Registry now supports reading custom backend services from a services.yaml file, simplifying custom configurations.

Reference: src/google/adk/cli/service_registry.py

12. Document experimental Daytona and E2B Environments

Doc file: docs/integrations/environment-toolset.md

Current state:

ADK provides a LocalEnvironment implementation for use with the Environment Toolset framework.

Proposed Change:

ADK provides a LocalEnvironment implementation for use with the Environment Toolset framework. ADK also provides experimental DaytonaEnvironment and E2BEnvironment remote sandbox environments.

Reasoning:
ADK added built-in implementations for DaytonaEnvironment and E2BEnvironment that extend BaseEnvironment.

Reference: src/google/adk/integrations/daytona/_daytona_environment.py

13. Add specialized Gemma 3 wrapper support

Doc file: docs/agents/models/google-gemma.md

Current state:

Focuses strictly on standard API/self-hosting without mention of Gemma 3's function calling workarounds.

Proposed Change:

Add a Gemma 3 Support section explaining that Gemma and Gemma3Ollama classes provide automatic workarounds for Gemma 3.

Reasoning:
ADK added specialized Gemma and Gemma3Ollama classes specifically to bridge the function calling and system instruction gaps for Gemma 3 models.

Reference: src/google/adk/models/gemma_llm.py

Other

14. Update CLI default models to gemini-3.5-flash

Doc file: docs/get-started/python.md, docs/agents/config.md

Current state:

model: gemini-flash-latest and model='gemini-flash-latest'

Proposed Change:

Update examples to model='gemini-3.5-flash' and model: gemini-3.5-flash.

Reasoning:
The adk create CLI tool now defaults to generating agent templates with gemini-3.5-flash instead of gemini-flash-latest.

Reference: src/google/adk/cli/cli_create.py

15. Remove deprecated flags from adk deploy agent_engine

Doc file: docs/integrations/bigquery-agent-analytics.md

Current state:

Mentions using --staging_bucket and --adk_app flags.

Proposed Change:

Remove --staging_bucket and --adk_app flags from the deployment examples.

Reasoning:
The --adk_app and --staging_bucket flags have been deprecated and removed from adk deploy agent_engine.

Reference: src/google/adk/cli/cli_deploy.py