Agent Connector: Use WorkItemPayload model for create_work_item_for_agent

[mikahanninen]

Summary

The payload parameter type dict is causing problems in tool discovery and recognition between the action server and the agent server.

Below is an example of agent reasoning, which eventually can cause unwanted disruption (manual input from the user) in the agent flow, which is meant to run automatically after initial input

Adapting to tool limitations

I've noticed that the API signature for functions.create_work_item_for_agent only requires agent_name, an attachments list, and the work_item_api_url. There's a mismatch between the tool's definition and what the Runbook Step 8 suggests. The tool doesn’t allow for payload fields, which limits what I can send.

Although the Runbook provides an extended signature, I need to follow the tool reference closely. I can still create a work item using the agent name and attachments, and I'll assume the context may be added automatically. After success, I’ll categorize the emails based on classification.

• Changed create_work_item_for_agent() payload parameter from dict to WorkItemPayload Pydantic model
• WorkItemPayload uses ConfigDict(extra="allow") for flexible payload structure
• Updated sema4ai-actions to 1.6.6
• Bumped version to 4.2.1

Test plan

• Test create_work_item_for_agent() with various payload structures
• Verify extra properties are passed through correctly

🤖 Generated with Claude Code

[raivolink]

LGTM