CleverThis Engineering
82 lines
3.5 KiB
Gherkin
82 lines
3.5 KiB
Gherkin
Feature: Tool Agent and Application Context Updates
|
|
As a developer using CleverAgents
|
|
I want context updates from tools and streams to be properly captured
|
|
So that state is maintained throughout agent execution
|
|
|
|
Background:
|
|
Given I have a configured CleverAgents application
|
|
|
|
Scenario: Tool agent preserves context updates in Python code
|
|
Given I have a tool agent with Python code
|
|
When the code updates the context dictionary
|
|
Then the context changes should be saved globally
|
|
And the changes should be available after execution
|
|
|
|
Scenario: Tool agent handles None context gracefully
|
|
Given I have a tool agent with Python code
|
|
When the code is executed without providing context
|
|
Then the execution should complete without errors
|
|
And no context updates should be saved
|
|
|
|
Scenario: Global context updates are collected from tool executions
|
|
Given I have multiple tool agents
|
|
When each tool updates different context keys
|
|
Then all context updates should be collected in _CONTEXT_UPDATES
|
|
And the updates should be merged into the global context
|
|
|
|
@skip
|
|
Scenario: Application collects context updates in single-shot mode
|
|
Given I have an application with context-aware agents
|
|
When I run the application in single-shot mode
|
|
Then context updates from all streams should be collected
|
|
And the global context should be updated with all changes
|
|
|
|
Scenario: StreamMessage preserves context reference when copied
|
|
Given I have a StreamMessage with context metadata
|
|
When the message is copied with modifications
|
|
Then the context reference should be preserved
|
|
And modifications to the context should affect the original
|
|
|
|
Scenario: Stream router maintains global context reference
|
|
Given I have a stream router with global context
|
|
When messages flow through the router
|
|
Then the global context reference should be maintained
|
|
And context updates should be applied to the same object
|
|
|
|
Scenario: Context updates with writing_stage are properly filtered
|
|
Given I have context updates with different writing_stage values
|
|
When the application processes the updates
|
|
Then only non-intro writing_stage updates should be merged
|
|
And intro stage updates should be ignored
|
|
|
|
Scenario: Interactive session saves context with context manager
|
|
Given I start an interactive session with a context manager
|
|
When I send messages and receive responses
|
|
Then each exchange should be saved to the context
|
|
And the conversation history should be preserved
|
|
|
|
Scenario: Context manager handles concurrent updates
|
|
Given I have multiple agents updating context simultaneously
|
|
When they all complete their updates
|
|
Then all updates should be captured without data loss
|
|
And the final context should contain all changes
|
|
|
|
Scenario: Tool agent exec preserves context object reference
|
|
Given I have a tool agent with code that modifies context in place
|
|
When the code executes with exec()
|
|
Then the original context object should be modified
|
|
And no deep copying should occur
|
|
|
|
Scenario: Application handles empty context updates gracefully
|
|
Given I have agents that may or may not update context
|
|
When some agents return empty context updates
|
|
Then only non-empty updates should be processed
|
|
And the application should not crash
|
|
|
|
@skip
|
|
Scenario: Context updates from different stream types are captured
|
|
Given I have input, processing, and output streams
|
|
When each stream type updates context
|
|
Then updates from all stream types should be collected
|
|
And no updates should be lost
|