Release/bridgic 0.4.0#193
Merged
Merged
Conversation
* first version of bridgic-cognitive
* Add test cases. Add the dependency of pyyaml
* Adjust the CognitiveContext to support dynamically adding the field summary for users.
* In AgentAutoma, add the context object initialization step. When using the context, there is no need to initialize it separately each time.
* Modify the Cognitive test cases to integration tests.
* Upgrade the ContextSkills interface. When necessary, override the add method yourself to automatically initialize the corresponding data structure. Adjust the package name, and expose all objects only through the __init__ module.
* Add a mechanism for setting the default Agent Automa initialization model as the default model for all thinking steps.
* Add some record information printing of the execution process.
* Adjust the capacity of the short-term memory for the historical record context from 5 to 20.
* The long-term memory compression logic for adjusting the historical record context has been updated, and the compression buffer has been increased to 10 entries. Memory compression is triggered only when the long-term memory buffer is full. Otherwise, these memories will be displayed as summaries, just like short-term memories.
* Adjust the log output length of _observation
* Fix the bug in AgentAutoma where the verbose parameter also controls verbose_prompt.
* Adjust the log output control of observation. The execution log will be output only when it is necessary, and the decision to output it will be based on the observation results.
* When selecting tools in the Default mode, add an observation step.
* Optimize the issue of repeatedly requested progressive disclosure information such as the already disclosed skills.
* Add the 'until' interface, which allows you to directly set the condition under which a thinking step should stop during runtime, and decouple the judgment condition from the context object at the same time.
* Let the thinking steps support the dynamic setting tools and skills.
* Let the context object determine which information is disclosed when the model is invoked and which is not.
* Adjust the context structure, remove the finish tag, and add the observation result and the status of each tool invocation. Add the default task termination judgment interface.
* Adjust prompt
* Adjust the prompt description to be consistent.
* Add data structure validation during the execution process of CognitveWorker.
* remove useless code
* Adapt the initialization of AgentAutoma to GraphAutoma.
* Allow the context type of AgentAutoma to be freely initialized.arun(goal=x) > ctx_init={goal: y} > context class default
* add test
* adjust arguments
* The goal of the context object can also be empty when it is adjusted. Fix the bug in ASLAutoma that does not retain the initialization parameters of custom Automa.
* Fix the bug where the AgentAutoma initialization context object fails to correctly identify field types.
* Adjust the context object initialization logic to be compatible with the direct initialization by passing either a list or an Exposure type.
* Feature/cognitive optimization (#183)
* Remove the Default mode
* Unify and abstract the observation and execution of thinking steps to a higher level, allowing the thinking steps to engage in pure reasoning.
* Reconstruct the prompt assembly logic for cognitive workers to make it more stable.
* Refactor cognitive framework: safety, naming, and prompt quality improvements.
* All the refactoring for the parts of cognitive that were not easy to use has been completed, and a large number of issues have been fixed.
* Adjust prompt design
---------
Co-authored-by: lynnlink <>
* Realize the replay function during the agent's execution process.
* Start to solve the system design problem of amphibious vehicles.
* Continue to optimize the system design of amphibious vehicles.
* Refactor the AgentAutoma module to make it well-organized and have a clear structure.
* Complete the design of the task structure marking during the agent's operation process.
* Adjust the code structure to further clarify the internal abstract relationships.
* Adjust the code for testing, fix the bugs, and then start the formal design of the workflow conversion operation.
* Add the implementation to the workflow and complete the code structure refactoring.
* Unify the parameter names of the sequential and loop tags. Adjust the entry method when running the workflow.
* Adjust the code structure; unify the log output.
* Adjust log
* Adjust the log output to be more structured to facilitate problem location.
* Adjust log
* fix workflow get empty action bug
* Adjust metadata and log
* Start to design loop workflow.
* Adjust the code structure of the workflow; Implement a feature that ensures the internal behavior of a specific model is not recorded. Implement dynamic degradation strategies for the workflow execution process.
* Debug the implementation of the loop to enable the loop condition and parameter updates to be implemented through code.
* Debug the implementation of the loop to enable the loop condition and parameter updates to be implemented through code.
* Reconstruct AgentAtuoma; The re-implemented AgentAutoma amphibious strategy
* Reorganize the code structure to make the current design of the cognitive module clearer.
* Adjust code
* Adjust test case
* Add a fallback plan: When the workflow encounters an error, it should be able to automatically switch back to the agent mode to solve the problem independently.
* Fix the bug where the framework swallows the exception thrown by the tool execution. Adjust the code structure
* Improve the trace information of the Agent execution process; Refactor the code structure and remove unnecessary classes.
* Adjust code
* first version of amphibipus agent system
* Refactor the naming of the bridgic-cognitive package and all its internal classes and methods; Add descriptor syntactic sugar when arranging the agent design pattern.
* Adjust the hook method of the thinking unit itself to the delegate design pattern; Make the declaration of steps in the workflow more concise.
* Add the capability of route selection for operation modes; Improve the design of the local degradation agent
* Create the bridgic-amphibious package
* Create the bridgic-amphibious package and add project configuration.
* Adjust doc-string and add Readme.md
* Adjust test case
---------
Co-authored-by: lynnlink <>
* first version of bridgic-cognitive
* Add test cases. Add the dependency of pyyaml
* Adjust the CognitiveContext to support dynamically adding the field summary for users.
* In AgentAutoma, add the context object initialization step. When using the context, there is no need to initialize it separately each time.
* Modify the Cognitive test cases to integration tests.
* Upgrade the ContextSkills interface. When necessary, override the add method yourself to automatically initialize the corresponding data structure. Adjust the package name, and expose all objects only through the __init__ module.
* Add a mechanism for setting the default Agent Automa initialization model as the default model for all thinking steps.
* Add some record information printing of the execution process.
* Adjust the capacity of the short-term memory for the historical record context from 5 to 20.
* The long-term memory compression logic for adjusting the historical record context has been updated, and the compression buffer has been increased to 10 entries. Memory compression is triggered only when the long-term memory buffer is full. Otherwise, these memories will be displayed as summaries, just like short-term memories.
* Adjust the log output length of _observation
* Fix the bug in AgentAutoma where the verbose parameter also controls verbose_prompt.
* Adjust the log output control of observation. The execution log will be output only when it is necessary, and the decision to output it will be based on the observation results.
* When selecting tools in the Default mode, add an observation step.
* Optimize the issue of repeatedly requested progressive disclosure information such as the already disclosed skills.
* Add the 'until' interface, which allows you to directly set the condition under which a thinking step should stop during runtime, and decouple the judgment condition from the context object at the same time.
* Let the thinking steps support the dynamic setting tools and skills.
* Let the context object determine which information is disclosed when the model is invoked and which is not.
* Adjust the context structure, remove the finish tag, and add the observation result and the status of each tool invocation. Add the default task termination judgment interface.
* Adjust prompt
* Adjust the prompt description to be consistent.
* Add data structure validation during the execution process of CognitveWorker.
* remove useless code
* Adapt the initialization of AgentAutoma to GraphAutoma.
* Allow the context type of AgentAutoma to be freely initialized.arun(goal=x) > ctx_init={goal: y} > context class default
* add test
* adjust arguments
* The goal of the context object can also be empty when it is adjusted. Fix the bug in ASLAutoma that does not retain the initialization parameters of custom Automa.
* Fix the bug where the AgentAutoma initialization context object fails to correctly identify field types.
* Adjust the context object initialization logic to be compatible with the direct initialization by passing either a list or an Exposure type.
* Feature/cognitive optimization (#183)
* Remove the Default mode
* Unify and abstract the observation and execution of thinking steps to a higher level, allowing the thinking steps to engage in pure reasoning.
* Reconstruct the prompt assembly logic for cognitive workers to make it more stable.
* Refactor cognitive framework: safety, naming, and prompt quality improvements.
* All the refactoring for the parts of cognitive that were not easy to use has been completed, and a large number of issues have been fixed.
* Adjust prompt design
---------
Co-authored-by: lynnlink <>
* Realize the replay function during the agent's execution process.
* Start to solve the system design problem of amphibious vehicles.
* Continue to optimize the system design of amphibious vehicles.
* Refactor the AgentAutoma module to make it well-organized and have a clear structure.
* Complete the design of the task structure marking during the agent's operation process.
* Adjust the code structure to further clarify the internal abstract relationships.
* Adjust the code for testing, fix the bugs, and then start the formal design of the workflow conversion operation.
* Add the implementation to the workflow and complete the code structure refactoring.
* Unify the parameter names of the sequential and loop tags. Adjust the entry method when running the workflow.
* Adjust the code structure; unify the log output.
* Adjust log
* Adjust the log output to be more structured to facilitate problem location.
* Adjust log
* fix workflow get empty action bug
* Adjust metadata and log
* Start to design loop workflow.
* Adjust the code structure of the workflow; Implement a feature that ensures the internal behavior of a specific model is not recorded. Implement dynamic degradation strategies for the workflow execution process.
* Debug the implementation of the loop to enable the loop condition and parameter updates to be implemented through code.
* Debug the implementation of the loop to enable the loop condition and parameter updates to be implemented through code.
* Reconstruct AgentAtuoma; The re-implemented AgentAutoma amphibious strategy
* Reorganize the code structure to make the current design of the cognitive module clearer.
* Adjust code
* Adjust test case
* Add a fallback plan: When the workflow encounters an error, it should be able to automatically switch back to the agent mode to solve the problem independently.
* Fix the bug where the framework swallows the exception thrown by the tool execution. Adjust the code structure
* Improve the trace information of the Agent execution process; Refactor the code structure and remove unnecessary classes.
* Adjust code
* first version of amphibipus agent system
* Refactor the naming of the bridgic-cognitive package and all its internal classes and methods; Add descriptor syntactic sugar when arranging the agent design pattern.
* Adjust the hook method of the thinking unit itself to the delegate design pattern; Make the declaration of steps in the workflow more concise.
* Add the capability of route selection for operation modes; Improve the design of the local degradation agent
* Create the bridgic-amphibious package
* Create the bridgic-amphibious package and add project configuration.
* Adjust doc-string and add Readme.md
* Adjust test case
* All tutorial files that have been added with the bridgic-amphibious module; The web page tutorial configuration is complete. Next, we'll start reviewing the content to ensure it runs smoothly.
* The final output form of optimizing the amphibious system: the agent uses the final summary as the result; for the workflow, the user can set the result by themselves.
* Complete quick start toturial
* Complete cognitive worker tutorial
* Adjust code
* Complete think tutorial
* Complete think tutorial
* Complete agent mode tutorial
* Complete context and exposure
* Complete custom otc (observaton-thinking-action) tutorial
* Complete cognitive policies tutorial
* Complete phase annotation tutorial
* Complete cognitive history tutorial
* Complete traceing tutorial
* first version bridgic-amphibious tutorial
* Adjust the output of the results. Fix the bug of parameter passing for the threshold of downgrade times. Add an after_action hook
* Complete API reference of birdgic-amphibious
---------
Co-authored-by: lynnlink <>
* first version of bridgic-cognitive
* Add test cases. Add the dependency of pyyaml
* Adjust the CognitiveContext to support dynamically adding the field summary for users.
* In AgentAutoma, add the context object initialization step. When using the context, there is no need to initialize it separately each time.
* Modify the Cognitive test cases to integration tests.
* Upgrade the ContextSkills interface. When necessary, override the add method yourself to automatically initialize the corresponding data structure. Adjust the package name, and expose all objects only through the __init__ module.
* Add a mechanism for setting the default Agent Automa initialization model as the default model for all thinking steps.
* Add some record information printing of the execution process.
* Adjust the capacity of the short-term memory for the historical record context from 5 to 20.
* The long-term memory compression logic for adjusting the historical record context has been updated, and the compression buffer has been increased to 10 entries. Memory compression is triggered only when the long-term memory buffer is full. Otherwise, these memories will be displayed as summaries, just like short-term memories.
* Adjust the log output length of _observation
* Fix the bug in AgentAutoma where the verbose parameter also controls verbose_prompt.
* Adjust the log output control of observation. The execution log will be output only when it is necessary, and the decision to output it will be based on the observation results.
* When selecting tools in the Default mode, add an observation step.
* Optimize the issue of repeatedly requested progressive disclosure information such as the already disclosed skills.
* Add the 'until' interface, which allows you to directly set the condition under which a thinking step should stop during runtime, and decouple the judgment condition from the context object at the same time.
* Let the thinking steps support the dynamic setting tools and skills.
* Let the context object determine which information is disclosed when the model is invoked and which is not.
* Adjust the context structure, remove the finish tag, and add the observation result and the status of each tool invocation. Add the default task termination judgment interface.
* Adjust prompt
* Adjust the prompt description to be consistent.
* Add data structure validation during the execution process of CognitveWorker.
* remove useless code
* Adapt the initialization of AgentAutoma to GraphAutoma.
* Allow the context type of AgentAutoma to be freely initialized.arun(goal=x) > ctx_init={goal: y} > context class default
* add test
* adjust arguments
* The goal of the context object can also be empty when it is adjusted. Fix the bug in ASLAutoma that does not retain the initialization parameters of custom Automa.
* Fix the bug where the AgentAutoma initialization context object fails to correctly identify field types.
* Adjust the context object initialization logic to be compatible with the direct initialization by passing either a list or an Exposure type.
* Feature/cognitive optimization (#183)
* Remove the Default mode
* Unify and abstract the observation and execution of thinking steps to a higher level, allowing the thinking steps to engage in pure reasoning.
* Reconstruct the prompt assembly logic for cognitive workers to make it more stable.
* Refactor cognitive framework: safety, naming, and prompt quality improvements.
* All the refactoring for the parts of cognitive that were not easy to use has been completed, and a large number of issues have been fixed.
* Adjust prompt design
---------
Co-authored-by: lynnlink <>
* Realize the replay function during the agent's execution process.
* Start to solve the system design problem of amphibious vehicles.
* Continue to optimize the system design of amphibious vehicles.
* Refactor the AgentAutoma module to make it well-organized and have a clear structure.
* Complete the design of the task structure marking during the agent's operation process.
* Adjust the code structure to further clarify the internal abstract relationships.
* Adjust the code for testing, fix the bugs, and then start the formal design of the workflow conversion operation.
* Add the implementation to the workflow and complete the code structure refactoring.
* Unify the parameter names of the sequential and loop tags. Adjust the entry method when running the workflow.
* Adjust the code structure; unify the log output.
* Adjust log
* Adjust the log output to be more structured to facilitate problem location.
* Adjust log
* fix workflow get empty action bug
* Adjust metadata and log
* Start to design loop workflow.
* Adjust the code structure of the workflow; Implement a feature that ensures the internal behavior of a specific model is not recorded. Implement dynamic degradation strategies for the workflow execution process.
* Debug the implementation of the loop to enable the loop condition and parameter updates to be implemented through code.
* Debug the implementation of the loop to enable the loop condition and parameter updates to be implemented through code.
* Reconstruct AgentAtuoma; The re-implemented AgentAutoma amphibious strategy
* Reorganize the code structure to make the current design of the cognitive module clearer.
* Adjust code
* Adjust test case
* Add a fallback plan: When the workflow encounters an error, it should be able to automatically switch back to the agent mode to solve the problem independently.
* Fix the bug where the framework swallows the exception thrown by the tool execution. Adjust the code structure
* Improve the trace information of the Agent execution process; Refactor the code structure and remove unnecessary classes.
* Adjust code
* first version of amphibipus agent system
* Refactor the naming of the bridgic-cognitive package and all its internal classes and methods; Add descriptor syntactic sugar when arranging the agent design pattern.
* Adjust the hook method of the thinking unit itself to the delegate design pattern; Make the declaration of steps in the workflow more concise.
* Add the capability of route selection for operation modes; Improve the design of the local degradation agent
* Create the bridgic-amphibious package
* Create the bridgic-amphibious package and add project configuration.
* Adjust doc-string and add Readme.md
* Adjust test case
* All tutorial files that have been added with the bridgic-amphibious module; The web page tutorial configuration is complete. Next, we'll start reviewing the content to ensure it runs smoothly.
* The final output form of optimizing the amphibious system: the agent uses the final summary as the result; for the workflow, the user can set the result by themselves.
* Complete quick start toturial
* Complete cognitive worker tutorial
* Adjust code
* Complete think tutorial
* Complete think tutorial
* Complete agent mode tutorial
* Complete context and exposure
* Complete custom otc (observaton-thinking-action) tutorial
* Complete cognitive policies tutorial
* Complete phase annotation tutorial
* Complete cognitive history tutorial
* Complete traceing tutorial
* first version bridgic-amphibious tutorial
* Adjust the output of the results. Fix the bug of parameter passing for the threshold of downgrade times. Add an after_action hook
* Complete API reference of birdgic-amphibious
* Add the bridgic-amphibious CLI scaffolding to create a project; Eliminate the redundant design for the dedicated intelligent agent as the pathfinding intelligent agent; Adjust the code structure
* Add bridgic-amphibious skill
* Adjust doc for removing sequential, loop
* Adjust the workflow to switch back to the agent mode and customize the context settings. By default, the context is clean.
* Update birdigic-amphibious skill for AgentCall
* Fix AgentCall history arguments bug
* Fix AgentAutoma arun bug when context and kwargs arguments pass
* Adjust the project structure and project code generated by the scaffolding.
* Reconstruct the type of objects generated in on_workflow to make the code more uniform and facilitate the maintenance and upgrade of on_workflow.
* Bridgic-amphibipus supports human-in-loop and awaits operation.
* Adjust doc
* Adjust bridgic-amphibious skills
* Adjust log level
* fix: align naming and tighten amphibious module
* refactor(amphibious): rename AMPHIFLOW, fix AUTO dispatch, relax LLM requirement
* refactor(amphibious): rename AMPHIFLOW, fix AUTO dispatch, relax LLM requirement
* Sync skills
* chore(skills): sync bridgic-amphibious, add bridgic-llms, support BRIDGIC_DEV_INDEX
* chore(skills): sync bridgic-amphibious, add bridgic-llms, support BRIDGIC_DEV_INDEX
* Adjust skill to ensure python environment to create bridgic-amphibious projrcy
* Adjust skill to ensure python environment to create bridgic-amphibious project
* Adjust skills
* Adjust skills
* Fix: Resolve the error in environment initialization
* Fix: Resolve the error in environment initialization
* Updata bridgic-amphibious skill .ini
* Adjust the order of tutorials
* Rename human_request_tool to request_human_tool (inner ask_human to request_human) for naming consistency, auto inject request_human_tool into every agent context tools during arun with name based dedupe so explicit passing still works and the LLM can call request_human in any mode without extra wiring, drop will_fallback from the public arun signature since AMPHIFLOW now always enables fallback and users should switch to RunMode WORKFLOW to opt out, freeze the builtin tools registry as an immutable tuple to prevent runtime mutation, and sync README tutorials skill docs and HITL tests including new coverage for builtin injection across agent workflow step fallback and full agent fallback paths.
* Remove redundant file
---------
Co-authored-by: lynnlink <>
NiceCode666
approved these changes
Apr 15, 2026
tsingfei
approved these changes
Apr 15, 2026
tielei
approved these changes
Apr 15, 2026
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
🚀 Bridgic v0.4.0 Release
Overview
This release introduces a brand-new integration module —
bridgic-amphibious— which brings cognitive amphibious agent capabilities to the Bridgic ecosystem. The mainbridgicpackage has also been updated to take this new module as a dependency.🛠️ Key Features
bridgic-amphibiousintegration package, providing cognitive amphibious agent capabilities (Feature/bridgic cognitvie amphibious #189)bridgic-amphibiousto streamline amphibious project creation (Feature/bridgic amphibious scaffolding #191)📚 Docs
bridgic-amphibious(Docs/bridgic amphibious #190)📦 Updated Packages
Added a new dependency of
bridgicso that it depends onbridgic-amphibious. The updated packages:bridgic==0.4.0bridgic-amphibious==0.1.0