- Refactored the initSandboxV2 function to return a structured result, consolidating the runner, computer, configuration, cleanup function, loading message ID, and roles into a single return type.
- Updated the Stream method to utilize the new sandboxV2InitResult structure, improving clarity and reducing complexity in handling sandbox initialization.
- Introduced role management enhancements, allowing for pre-resolved role connectors to be passed through the request, streamlining connector resolution during execution.
- Adjusted various components to support the new roles structure, ensuring consistent handling across the agent's sandbox operations.
- Added logging for connector resolution and role management, improving diagnostics and traceability during sandbox execution.
- Updated .gitignore to include tools/TOOL-REGISTRATION.md for better project organization.
- Added support for tool loop processing when tool call responses are present and sandbox mode is disabled, improving the assistant's ability to handle complex tool interactions.
- Implemented fallback delegation to a loop fallback mechanism in case of tool loop execution failure, enhancing robustness in error scenarios.
- Updated the assistant message structure to include reasoning content, providing better context for generated responses.
- Enhanced error logging in tool call execution to include detailed content, improving diagnostics for tool call failures.
- Updated system configuration to include a new loop fallback agent, expanding the assistant's capabilities.
- Integrated logging for connector resolution failures, providing clearer diagnostics during fallback scenarios.
- Simplified connector ID extraction in various components, ensuring accurate handling of model identifiers.
- Updated model configuration to support new parameters and capabilities, improving overall provider management.
- Enhanced OpenAPI settings to reflect changes in model options and connector behavior, ensuring better alignment with upstream API requirements.
- Updated the system configuration to include new role-level defaults for Light, Vision, and Audio connectors.
- Refactored the resolveSystemConnector function to prioritize per-agent overrides, improving connector resolution logic.
- Enhanced LLMConnector integration across various components to streamline settings retrieval and capabilities management.
- Improved error handling and logging for connector-related operations, ensuring better diagnostics and user feedback.
- Replaced 'Voice' with 'Audio' in the system configuration and related tests to better reflect functionality.
- Introduced new methods for role management in the llmprovider, allowing for dynamic retrieval of roles based on user and team context.
- Updated the OpenAPI settings to support new role management endpoints and capabilities.
- Enhanced the handling of API keys in provider management, allowing for optional plain-text retrieval.
Modified the initSandboxV2 function to return a per-request copy of the SandboxConfig, ensuring that concurrent requests to the same assistant have their own mutable configuration. Updated related logging and cleanup processes to utilize the new configuration structure. Additionally, introduced a DefaultOneShotMaxAge constant to manage the lifecycle of OneShot containers more effectively.
- Introduced a new sandboxV2StreamParams struct to group parameters for the executeSandboxV2Stream function, improving code clarity and maintainability.
- Updated the GetComputer function to handle connector configuration injection via the a2o proxy, removing the need for direct connector parameters.
- Enhanced the executeSandboxV2Stream function to utilize the new parameters struct, simplifying the function signature and improving readability.
- Implemented RegisterProxyConfigs to inject OpenAI-compatible connector configurations into the a2o proxy, ensuring proper environment setup for sandbox execution.
- Refactored BuildCreateOptions to remove direct connector handling, aligning with the new configuration injection approach.
- Added detailed trace logging in the Assistant's Stream function to capture computer and workspace information.
- Replaced fmt.Printf statements with kunlog.Trace in the Telegram polling adapter for consistent logging.
- Updated sandbox lifecycle management functions to utilize kunlog for logging workspace and node resolution details.
- Enhanced error handling in the Claude parser with trace logging for JSON unmarshal and scanner errors.
- Improved file handling logging in the workspace API to provide clearer insights into file read operations.
- Added support for "execute" message type to track tool execution observations within the sandbox environment.
- Implemented handling for execute messages in the stream processing, allowing for real-time updates on tool execution status.
- Enhanced the stream parser to manage the lifecycle of execute messages, including merging input and output data.
- Updated the stream handler to accommodate new message types, improving overall message processing capabilities.
- Refactored related tests to ensure coverage for the new execute message functionality, enhancing reliability.
- Implemented a new endpoint to retrieve the absolute path of a workspace's root directory.
- Enhanced the workspace interface with a GetRoot method to facilitate this functionality.
- Updated the workspace manager to utilize the new method for improved path resolution.
Made-with: Cursor
- Implemented V2 sandbox initialization in the assistant loading process, allowing for standalone sandbox.yao configuration.
- Added support for V2 sandbox execution paths in the Assistant's Stream method, differentiating between V1 and V2 sandboxes.
- Introduced comprehensive tests for V2 sandbox configurations, ensuring correct loading and execution behavior.
- Updated the context and types to accommodate V2 sandbox features, including system information and workspace management.
Made-with: Cursor
- Introduce a fallback mechanism in the GetConnector method to use a default connector if the specified connector is not found.
- Add logging to notify when falling back to the default connector, improving debugging and traceability.
- Update comments for clarity on connector selection priority, ensuring better understanding of the logic flow.
- Remove the `LoadModelCapabilities` test and associated model capabilities initialization from the agent, streamlining the loading process.
- Update the LLM provider implementations to utilize a unified `Capabilities` structure, replacing references to `openai.Capabilities` with `llm.Capabilities`.
- Enhance capability retrieval methods to simplify the extraction of connector capabilities, ensuring compatibility across different LLM providers.
- Clean up unused functions and variables related to model capabilities, improving code maintainability.
- Add new indirect dependencies including various Charmbracelet packages for improved UI handling.
- Enhance logging in the Assistant module by adding tool completion and start logging for better traceability of tool calls.
- Modify context handling in the RequestLogger to support a stack-based assistant ID management, improving the logging structure for agent requests.
- Implement event service integration for better trace management and debugging capabilities.
- Add a new `MergeMetadata` function in the `Context` struct to allow merging of caller-provided metadata into the context, enabling sub-agent hooks to access this information.
- Update the `Stream` method in the `Assistant` to utilize the new metadata merging functionality, enhancing the context management for sub-agents.
- Implement secure config file location (/tmp/.yao/proxy.json) instead of user-visible /workspace/
- Add generic options map support for backend-specific parameters (e.g., thinking for Volcengine GLM-4.7)
- Add secrets support for passing sensitive env vars (e.g., GITHUB_TOKEN) to sandbox container
- Remove excessive debug logs, keep critical ones with log.Printf("[Sandbox]...")
- Fix test assertions for system prompt passing via CLI args instead of env var
- Update i18n messages for sandbox loading states
Co-authored-by: Cursor <cursoragent@cursor.com>
- Added functionality to build and manage MCP configuration for sandbox environments, allowing for dynamic tool execution.
- Enhanced the Assistant's Stream method to skip MCP tool calls in sandbox mode, with internal handling by Claude CLI.
- Introduced unit tests for MCP configuration building and skills directory resolution, ensuring robust integration.
- Updated sandbox manager to create IPC sessions and manage tool exposure dynamically, improving interaction with external agents.
- Enhanced documentation to reflect new features and integration points for MCP and skills within the sandbox.
- Added steps to pull necessary Docker images for sandbox testing in both CI workflows.
- Updated the AI test execution to utilize sandbox configurations, ensuring proper environment setup.
- Introduced sandbox initialization in the Assistant's Stream method, allowing for execution of coding agents like Claude and Cursor.
- Enhanced context management to support sandbox execution, improving flexibility in handling agent operations.
- Introduced mechanisms to handle agent-to-agent (A2A) calls, including automatic history skipping for forked calls and proper source tracking.
- Enhanced context management with the addition of ForkParentInfo to facilitate child stack creation without race conditions.
- Updated JSAPI methods to ensure correct handling of sub-agent calls, maintaining output isolation and preventing history pollution.
- Improved documentation to clarify the behavior of A2A calls and context management in concurrent scenarios.
- Updated the Stream method to include options for handling message history and context more effectively, ensuring original messages are preserved for autoSearch and delegation.
- Introduced a new buildContextMessage function to consolidate conversation context, filtering out system messages and limiting to the last five user messages for efficiency.
- Enhanced content processing by adding a convertToContentParts function to handle different content formats, improving compatibility with historical data.
- Improved logging and error handling in the executeLLMStream method to ensure clarity in LLM request tracing and response handling.
- Added new utility functions for extracting text content and building context messages, enhancing overall code clarity and maintainability.
- Updated the BufferUserInput method to ensure only the root stack buffers user input, preventing duplication in delegated agents.
- Added comments in the Stream and processNextResponse methods to clarify that user input is already buffered by the root agent, allowing delegated agents to skip this step.
- Improved code clarity and maintainability by documenting delegation behavior in the context of user input handling.
- Introduced delegation functionality in the Create hook, allowing agents to route requests to sub-agents without invoking LLM processing.
- Updated the HookCreateResponse structure to include a Delegate field, enabling immediate delegation to another agent.
- Enhanced the Stream method to handle delegation responses, ensuring proper stream closure and error handling for delegated agents.
- Improved logging for delegation actions to facilitate debugging and traceability in agent interactions.
- Added support for `tool_called` and `tool_result` assertions in the Asserter, allowing for validation of tool execution and arguments.
- Introduced methods to check if specific tools were called and to validate their results against expected patterns.
- Updated the `README.md` to include detailed documentation on new assertion types, including usage examples and value formats.
- Enhanced the dynamic runner to set the response for tool-related assertions, improving the overall testing framework's capabilities.
- Updated the search handling to incorporate keyword extraction with weights, improving the relevance of search results.
- Refactored the `shouldAutoSearch` method to return a `SearchIntent` struct, allowing for more nuanced control over search execution based on context.
- Enhanced the `buildSearchRequests` function to utilize extracted keywords, optimizing search queries based on user input.
- Improved the handling of search types and conditions, ensuring that the system can dynamically adjust search behavior based on intent and configuration.
- Updated documentation and prompts to reflect changes in keyword extraction and search intent classification, providing clearer guidelines for usage.
- Updated the agent's Stream and response processing methods to return and handle *context.Response directly, eliminating the need for type assertions.
- Simplified test cases by removing unnecessary type conversions and directly accessing response fields.
- Enhanced the extraction of data from Next hook responses, ensuring more robust handling of custom data structures.
- Improved overall code readability and maintainability by streamlining response handling logic across various components.
- Renamed several assistant packages for clarity, including "Entity Extraction" to "Entity Extractor" and "Keyword Extraction" to "Keyword Extractor."
- Revised descriptions for various assistants to enhance understanding of their functionalities, such as changing "Extract keywords from text content" to "Extract search keywords."
- Added a "uses" field with "search" set to "disabled" in the configuration of each assistant, standardizing their setup.
- Updated the "Prompt Optimizer" description to "Optimize prompts for better results" and modified the "QueryDSL Generator" to "Query Builder" for improved clarity.
- Ensured consistent naming conventions and descriptions across all assistant packages to enhance user experience and documentation clarity.
- Updated the `shouldAutoSearch` method to include additional parameters for improved intent detection, allowing for better decision-making on whether to execute auto search.
- Introduced a new `checkSearchIntent` method to utilize the `__yao.needsearch` agent for determining the necessity of a search based on user input.
- Implemented a `ClearExcept` method in the cache to selectively clear non-system agents while preserving essential system agents during cache management.
- Updated the `LoadBuiltIn` function to maintain system agents in the cache, ensuring they remain available for use.
- Enhanced test coverage for loading system agents and validating search intent detection, ensuring robustness in the assistant's search capabilities.
- Revised localization files to include new messages for search intent feedback, improving user experience during search operations.
- Updated the executeAutoSearch method to include an optional parameter for Skip.Keyword, allowing for conditional keyword extraction during web searches.
- Implemented logic to extract keywords only when configured and not skipped, improving search query optimization.
- Modified the Assistant's Stream method to pass options to executeAutoSearch, ensuring seamless integration of the new functionality.
- Updated DESIGN.md to document the changes in keyword extraction logic and its impact on the search process.
- Added functionality to the Assistant's Stream method to execute auto search if enabled, enhancing the search capabilities based on user configuration.
- Introduced helper methods for determining auto search eligibility, executing the search, and injecting search context into messages.
- Updated DESIGN.md to reflect the new auto search logic, including detailed descriptions of the new methods and their integration points within the search process.
- Added logging for hook start and completion in the Assistant's Stream method to improve traceability during execution.
- Refactored the AgentGetterFunc to utilize the caller package, enhancing modularity and reducing circular dependencies.
- Updated the CallAgent function to check for the initialized AgentGetterFunc from the caller package, ensuring proper agent loading.
- Enhanced the Search handler to support an optional context parameter, improving flexibility for agent mode operations.
- Refined the agentSearch function to delegate search requests to a new AgentProvider, streamlining the search process.
- Updated DESIGN.md to reflect changes in search modes and the integration of the caller package, ensuring comprehensive documentation.
- Added a new method to set store settings during assistant initialization, allowing for configuration of storage parameters such as MaxSize and TTL.
- Updated context creation methods to streamline the setup process, ensuring that essential fields are populated consistently across various test contexts.
- Revised tests to validate the new initialization behavior and context management, ensuring proper handling of assistant settings and context properties.
- Introduced methods for initializing and managing a chat buffer, allowing for efficient storage of user inputs and assistant messages during chat sessions.
- Added functionality to track execution steps, including beginning and completing steps, with support for capturing space snapshots and handling errors.
- Enhanced the FlushBuffer method to save buffered messages and steps to the database, ensuring data integrity and recovery capabilities.
- Updated the Stream method to integrate buffer management, ensuring proper handling of chat sessions and message storage.
- Added comprehensive tests to validate buffer initialization, user input handling, and step tracking functionalities.
- Added functionality to load Knowledge Base (KB) configuration from `agent/kb.yml`, allowing dynamic settings for chat sessions.
- Introduced `initKBConfig` function to read and parse KB settings, integrating them into the Assistant's initialization process.
- Enhanced the Assistant's conversation initialization to prepare KB collections asynchronously, improving performance during chat interactions.
- Updated tests to verify the correct loading and application of KB settings, ensuring robust integration with the Assistant's functionality.
- Refactored metadata handling for KB collections to include additional fields for improved context management during chat sessions.
- Added permission validation in the Assistant's Stream method to ensure user authorization before processing input messages.
- Introduced conversation initialization within the Stream method to prepare the context for chat interactions.
- Removed the history.go file, which previously contained a placeholder method for handling chat history, streamlining the Assistant's codebase.
- Updated the Knowledge Base API integration in collection management, ensuring all collection operations utilize the new API structure for improved consistency and error handling.
- Added a new GetDel method to the space object, allowing retrieval and immediate deletion of a key's value, enhancing one-time use data handling.
- Updated the JS API to include tests for GetDel functionality, ensuring correct behavior for existing and non-existent keys, as well as complex data structures.
- Removed debug print statements from the Assistant's initializeCapabilities method for cleaner code.
- Replaced all instances of `Script` with `HookScript` in the Assistant and related files to improve clarity and consistency in naming.
- Updated method calls in the Stream, Create, and Next hooks to utilize the new `HookScript` field.
- Adjusted tests and benchmarks to reflect the changes in script handling, ensuring all functionalities remain intact and operational.
- Enhanced the load and initialization processes to accommodate the new HookScript structure, streamlining the assistant's script management.
- Updated the Stream, BuildContent, and LLM execution methods to accept an Options parameter, allowing for more flexible context handling.
- Removed debug print statements to clean up the code and improve readability.
- Enhanced locale handling in the loadMap function to automatically inject assistant name and description into all locales, ensuring better localization support.
- Introduced output skipping functionality in context options to manage internal A2A calls more effectively.
- Improved output writer resolution logic to prioritize context settings, enhancing output management during agent calls.
- Updated the Assistant methods to accept an Options parameter, enhancing flexibility in context management.
- Removed the Connector field from the context and related structures, transitioning to a more streamlined options-based approach.
- Adjusted various tests to accommodate the new options handling, ensuring comprehensive coverage of the updated functionality.
- Enhanced the Create and Next hooks to return options alongside responses, improving the overall usability of the API.
- Cleaned up deprecated fields and improved context initialization for better maintainability.
- Added support for converting extended types (file, data) to standard LLM types (text, image_url, input_audio) in the Stream method.
- Introduced a new agentCallerWrapper to facilitate agent calls from the content package.
- Implemented ThreadID management for nested agent calls to improve concurrent stream identification.
- Enhanced message handling to include metadata for message_start and message_end events, allowing for better tracking of message states.
- Removed deprecated vision capability checks from the Assistant initialization process, streamlining the codebase.
- Added new content types (file, data) to the context types for improved message content handling.
- Replaced the deprecated getConnectorCapabilities method with a unified capability getter in the Assistant model, improving capability retrieval logic.
- Added a new API endpoint to retrieve essential assistant information, including fields like id, name, avatar, and connector options, enhancing the assistant's data accessibility.
- Updated the LLM management to support filtering by capabilities, allowing for more flexible provider listings based on user-defined models.
- Improved overall structure and clarity in the assistant's capabilities and API responses, ensuring better maintainability and usability.
- Enhanced the getConnectorCapabilities method to prioritize model capabilities and connector settings, improving capability retrieval logic.
- Deprecated the tools field in the Assistant model, transitioning to MCP for tool management, and updated related methods accordingly.
- Introduced new fields for connector options and prompt presets in the Assistant model, allowing for more flexible configurations.
- Updated the GetAssistant method to support field selection, improving data retrieval efficiency and flexibility.
- Refactored tests and documentation to reflect changes in the assistant structure and capabilities, ensuring clarity and maintainability.
- Updated the model capabilities throughout the agent to utilize the new gouOpenAI.Capabilities struct instead of the previous ModelCapabilities.
- Adjusted related methods and types to ensure compatibility with the new capabilities structure, enhancing clarity and maintainability.
- Improved context handling and message processing by directly integrating OpenAI capabilities, streamlining the overall architecture.
- Updated the traceAgentOutput method to be deprecated in favor of the new traceAgentCompletion method, which creates a dedicated completion node for reporting final outputs.
- Enhanced the i18n messages to include labels and descriptions for the new agent completion functionality, improving localization support.
- Removed debug logging related to MCP tools from the buildCompletionOptions method to streamline the code and improve clarity.
- Updated the Stream method to return a flexible final response, accommodating both standard and Next hook responses.
- Introduced NextProcessContext to encapsulate context for processing Next hook responses, enhancing clarity and maintainability.
- Removed deprecated Done and Failback hooks, streamlining the hook management process.
- Enhanced error handling and logging for improved traceability during streaming operations.
- Added detailed debug logging throughout the Stream method to trace execution flow and tool call results, improving visibility during runtime.
- Implemented error tracing with the new traceAgentFail method to capture failures in agent nodes, enhancing error handling.
- Updated MCP tools debugging to provide insights into tool application and validation processes, aiding in troubleshooting.
- Refactored output handling in traceAgentOutput for better clarity and consistency in response management.
- Renamed methods in ToolCallResult from ServerID to Server and ToolName to Tool for improved clarity and consistency.
- Updated i18n messages in builtin.go to remove redundant context, simplifying the label and description for the assistant's streaming process.
- Enhanced the Stream method to include a retry mechanism for tool calls, allowing for up to three attempts in case of errors.
- Introduced detailed logging for each retry attempt, improving traceability and debugging capabilities.
- Added a new method, buildToolRetryMessages, to construct messages for LLM retries, ensuring proper context is maintained.
- Updated the executeLLMForToolRetry method to handle streaming output during retries, providing real-time feedback to users.
- Improved error handling to differentiate between retryable and non-retryable errors, enhancing the robustness of tool call executions.
- Refactored the BuildRequest method to improve clarity and maintainability by separating the building of completion options and final messages.
- Updated the buildMessages function to include MCP samples as system messages, enhancing the message prioritization logic.
- Introduced applyMCPTools method to integrate MCP tools into completion options, improving functionality and flexibility in tool usage.
- Enhanced error handling in completion options to ensure robust application flow.
- Updated MCP server configuration to support advanced formats, including tools and resources.
- Refactored the Assistant's Stream method for improved readability and maintainability, adding clear section comments.
- Enhanced error handling and logging for better traceability during streaming operations.
- Removed deprecated methods and streamlined MCP server handling in tests to ensure robust functionality and validation.
- Simplified the Stream method by extracting capability initialization into a separate function, improving readability and maintainability.
- Enhanced error handling by centralizing trace logging for failures and history tracking.
- Removed the MCP hook implementation as it was no longer needed, streamlining the codebase.
- Updated agent trace handling to ensure consistent logging of agent output and errors.
