Minimal Claude Agent SDK support

[jasonyang101]

No description provided.

[socket-security]

Review the following changes in direct dependencies. Learn more about Socket for GitHub.

Diff	Package	Supply Chain Security	Vulnerability	Quality	Maintenance	License
	agentex-sdk@​0.6.5 ⏵ 0.6.7	-13
	anthropic@​0.74.0
	docstring-parser@​0.17.0
	claude-agent-sdk@​0.1.8

View full report

[danielmillerp]

Great job getting this together, my only feedback here is that we don't have a way of creating activities as tools since you have wrapped the whole agent loop as an activity. This is important because it enables us to do human in the loop and have idempotent tools. Maybe we can talk to Temporal team for a V1 version?

[prassanna-ravishankar]

Great job getting this together, my only feedback here is that we don't have a way of creating activities as tools since you have wrapped the whole agent loop as an activity. This is important because it enables us to do human in the loop and have idempotent tools. Maybe we can talk to Temporal team for a V1 version?

Regarding this, basically claude agent sdk doesn't offer too much control on what goes on within an agent call (thats offloaded to the agent). Basically the approach here has been to run the loop within an activity. But the loop itself just runs for one turn (maybe a multi-step turn). And then returns back with a session id.

And when we make the next call "we resume" the session with the id and based on the user prompt. This allows us to do human in the loop.