Dmitry Pasiukevich

This commit performs a significant **refactoring** by **renaming the request and response types for memory search operations**. Specifically, `SearchMemoryRequest` is now `SearchRequest` and `SearchMemoryResponse` is now `SearchResponse`. This **breaking change** impacts the **core `memory` subsystem**, including the `SearchMemory` method signature across `memory/service.go`, `memory/inmemory.go`, and `internal/memory/memory.go`. The change propagates to the **`agent` and `tool` contexts** that interact with memory, requiring updates to their `SearchMemory` method signatures and associated test files. This improves API consistency and clarity, but requires downstream consumers to update their type references.

This commit **refactors** the **`adkrest`** package to **remove its direct dependency on the `launcher` package**, enhancing modularity and preparing for future `launcher` revisions. The `server/adkrest/handler.go` file now exposes a new `Server` struct and `NewServer` constructor, replacing the previous `NewHandler` function, which represents a **breaking API change**. Consequently, the `cmd/launcher/web/api/api.go` module has been updated to utilize this new `adkrest.NewServer` constructor with explicit configuration, including telemetry processors. This ensures critical server packages no longer implicitly rely on `launcher` for configuration.

This commit introduces a significant **refactoring** of the **memory service** to achieve consistent naming conventions across the ADK, preparing for the upcoming v1.0 release. Key changes include renaming core methods such as `AddSession` to `AddSessionToMemory` and `Search` to `SearchMemory` within the `agent`, `memory`, and `tool` modules, alongside updates to their return types and associated data structures like `SearchRequest`/`SearchResponse`. Furthermore, the `Entry` struct in the `memory` service now includes `ID` and `CustomMetadata` fields, enhancing the data stored for each memory entry. This constitutes a **breaking change** for any downstream consumers directly utilizing the `memory` service's public interface, necessitating updates to method calls and type definitions to align with the new API.

This commit primarily **refactors the `tool` package** by **removing a duplicated `ConfirmationProvider` type** from the `mcptoolset` module and consolidating its definition into the shared `tool` package. This **code consolidation** standardizes the `ConfirmationProvider` across the codebase, requiring all references within `mcptoolset` to now use `tool.ConfirmationProvider`. Additionally, the commit **improves predicate handling** within the `tool` package by introducing `AllowedToolsPredicate` and deprecating `StringPredicate`, alongside a minor structural cleanup in `exitlooptool`. This change is a **breaking change** for any external code directly referencing the now-removed `mcptoolset.ConfirmationProvider`, promoting better maintainability and consistency.

This commit **introduces a new Apigee LLM model** implementation, significantly expanding the system's AI integration capabilities. The core **new capability** is provided in `model/apigee/apigee.go`, offering content generation and configuration options such as `WithProxyURL` and `WithCustomHeaders`. Supporting this, the `LLMAgent`'s initialization logic in `agent/llmagent/llmagent.go` is **refactored** to correctly handle agent types, and a `TypeRemoteAgent` constant is added to `internal/agent/state.go`. Dedicated unit tests for the Apigee model are included, alongside minor unrelated changes like an improved error message in `session/inmemory.go` and a typo fix in `agent/context.go`.

This commit performs a **maintenance chore** by updating the **internal version constant** within the `internal/version/version.go` file. Specifically, the version string has been incremented from `0.3.0` to `0.6.0`. This change primarily affects the project's **internal versioning system**, ensuring that any component relying on this constant reflects the latest software release state.

This commit **fixes** an issue in the **`mcptoolset` client** by updating its error handling logic for session-related problems. Specifically, it modifies the `shouldRefreshConnection` function in `tool/mcptoolset/client.go` to recognize the new `mcp.ErrSessionMissing` error type. This **maintenance update** replaces a deprecated string-based check for session not found, improving the reliability and clarity of connection refresh decisions. The change ensures the client correctly identifies when a session is missing, leading to more robust connection management within the `mcptoolset` subsystem.

This commit performs **test maintenance** by updating an expected model name within the **Gemini model integration tests**. Specifically, it corrects the expected model version string in `model/gemini/gemini_test.go` from `gemini-2.0-flash` to `gemini-2.5-flash`. This **test fix** ensures that the test suite accurately reflects the current **Gemini API model names**, preventing outdated expectations and potential false test failures. The change is limited to the test suite and has no direct impact on the production code's functionality.

This commit **updates and regenerates test fixtures** for the **`ParallelAgent` workflow agent** to incorporate recent changes, specifically the addition of an `identityProcessor`. It involves updating the `modelName` constant in `agent/workflowagents/parallelagent/agent_test.go` from `gemini-2.0-flash-exp` to `gemini-2.5-flash`. Additionally, the recorded HTTP requests and responses in `TestParallelAgentWithTools_agent1.httprr` and `TestParallelAgentWithTools_agent2.httprr` are refreshed to reflect the new model and updated system instructions. This **test maintenance** ensures the **integration tests** accurately validate the `ParallelAgent`'s behavior with the latest configurations.

This commit performs a **cleanup and refactoring** by completely removing the unused `MutableSession` struct from the codebase, as it was a leftover from previous refactoring and served no functional purpose. This **maintenance** task primarily impacts the **session management** logic within the `runner` module, where the `Run` method now directly uses the stored session instead of an unnecessary wrapper. Additionally, various **test files** across `internal/llminternal`, `internal/memory`, and `tool/agenttool` are updated to remove references to the deprecated struct, ensuring consistency. The change streamlines internal session handling, improving code clarity and reducing unnecessary abstraction.

This commit introduces a **fix** and **refactoring** to the **LLM internal processing logic**, specifically for **Google Gemini models**. It refactors the **LLM variant detection** within `internal/llminternal/googlellm` by introducing a new `GoogleLLM` interface and functions like `NeedsOutputSchemaProcessor` to accurately determine if a Gemini model, particularly **Gemini 2.5 and below API variants**, requires an output schema processor. This ensures that display names are correctly handled in file uploads and that the output schema processor is appropriately enabled for older Gemini versions, preventing potential issues with model output interpretation. The `model/gemini` module is updated to expose the necessary variant information, improving the robustness of model interaction.

This commit **fixes** an issue within the **LLM internal event generation** module by ensuring that `long_running_function_ids` are correctly passed in the event generated for **request confirmation**. It assigns a unique ID to the request confirmation function call and streamlines content creation within the `generateRequestConfirmationEvent` function. As part of this **bug fix** and **refactoring**, the logic for generating function call IDs was extracted into a new utility function, `GenerateFunctionCallID`, in `internal/utils/utils.go`. New unit tests were also added to `internal/llminternal/functions_test.go` to cover various confirmation scenarios and validate the correct behavior of `generateRequestConfirmationEvent`.

This commit introduces a **new feature** by adding a dedicated **logging plugin** to the system. Implemented in `plugin/loggingplugin/logging_plugin.go`, this plugin provides detailed console output for various lifecycle events across **agent**, **model**, and **tool** interactions. It leverages methods like `onUserMessage`, `beforeRun`, `afterRun`, `beforeAgent`, `afterAgent`, `beforeModel`, `afterModel`, `beforeTool`, and `afterTool` to capture and log operational flow. This enhancement significantly improves observability and debugging capabilities for core components by providing granular insights into their execution.

This commit primarily **updates documentation** across the project, enhancing clarity for key configuration parameters. Specifically, it adds detailed explanations for `Instruction` and `InstructionProvider` within the **LLM agent configuration** and for `ProjectID`, `Location`, and `ReasoningEngine` in the **VertexAI service configuration**. Additionally, a minor **refactoring** was performed in `session/vertexai/vertexai_client.go` to rename an internal helper function `IsNotFoundError` to `isNotFoundError` for consistency. This work improves the overall maintainability and user understanding of critical system settings related to **LLM agent templating** and **VertexAI integration**.

This commit **refactors** the application's plugin management to consistently pass a complete `runner.PluginConfig` object instead of only a list of plugins. This **fixes** an underlying issue where partial plugin information was being propagated, ensuring that all necessary configuration details are available for proper plugin initialization. The change impacts the **launcher's console and A2A setup** (`cmd/launcher/*`) and the **ADK REST API's runtime controller** (`server/adkrest/controllers/runtime.go`), improving the robustness and consistency of how plugins are configured and managed throughout the system.

This commit introduces a **new capability** to the **`tool` package** by adding the `FilterToolset` helper function. This function, along with its supporting `filteredToolset` struct and methods, provides a dedicated and more robust mechanism for **filtering toolsets**. Concurrently, the `ToolFilter` field in the `Config` struct within `tool/mcptoolset/set.go` has been **deprecated**, advising users to migrate to the new `tool.FilterToolset` for improved toolset management. This **refactoring** streamlines how toolsets are filtered, offering a more explicit and maintainable approach for consumers of the `tool` package.

This commit delivers a **bug fix** for the **web server launcher** within the `cmd/launcher/web` package. It modifies the `Run` function to explicitly return `nil` upon a successful server shutdown, rather than propagating a context error, which ensures that callers correctly interpret a clean termination. This change improves the robustness and clarity of the **server lifecycle management**. Additionally, a minor **maintenance** update was applied to the `.github/workflows/go.yml` CI configuration, enabling race detection and test shuffling for enhanced reliability.

This commit **fixes a bug** in the **`agenttool`** that prevented it from properly surfacing errors originating from sub-agents, particularly those using `error_code` or `error_message` fields instead of standard Go errors. Previously, the `Run` function in `tool/agenttool/agent_tool.go` would return an empty response, effectively hiding these critical errors when wrapping `remoteagent` or similar components. The change introduces new error handling logic within `agenttool` to correctly detect and propagate these specific error types. This ensures that **`agenttool` now accurately reports all sub-agent errors**, significantly improving observability and debugging for users. A minor **style fix** was also applied to an error message in `cmd/launcher/web/web.go`.

This commit introduces a **bug fix** to the **database session service** by correcting how event timestamps are truncated before being stored. Specifically, the `AppendEvent` function in `session/database/service.go` is updated to use `time.Truncate(time.Microsecond)` while preserving the original time's location (timezone information). This ensures that event timestamps are accurately stored with their correct context, matching the database's maximum precision without losing critical metadata. The change primarily **fixes test failures** related to time zone discrepancies, improving the reliability of event handling within the **session management subsystem**.

This commit **updates the project's contributor guidelines** by clarifying policies within the **Contributor License Agreement (CLA)** section of `CONTRIBUTING.md`. This **documentation update** specifically addresses the use of **coding assistance tools** and **AI-generated code**, providing essential guidance for future contributions. The change ensures that all contributors understand the project's stance on these modern development practices, streamlining the contribution process and maintaining legal clarity.