Back to catalog / Agent runtime

Run a one-shot agent task with query

Run one task through the async message stream and distinguish text blocks from its final result.

Goal

Complete an auditable one-shot task with query() and inspect its terminal result.

Install the Python SDK and import query, AssistantMessage, TextBlock, and ResultMessage. [readme]
Use async for over messages from query(prompt=..., options=...), collecting text only from TextBlock items in AssistantMessage. [readme]
Record ResultMessage.subtype, result, and total_cost_usd; do not treat an intermediate message as success. [readme]

Confirm a terminal ResultMessage and associate its result and cost fields with the task ID.

If no terminal message arrives, check for async-iteration cancellation or CLI disconnection; do not infer success from the last text block.

Python 3.10+, claude-agent-sdk, and a supported Claude Code authentication setup.

Submit only the scoped task prompt; store output only in a caller-approved log location.

Model output still needs application validation; local SDK calls require supported Claude Code authentication and CLI runtime.

These values describe the upstream repository, not popularity or likes for an individual skill or playbook.

No metrics snapshot is available yet.

License: MIT