October 28, 2026: Cortex Agents Coding Agent thinking effort, system instructions, and stage mounts¶

The Cortex Agents Coding Agent agent:run API now accepts three additional public controls on a request that includes the code_toolset_all tool:

• thinking_effort sets the orchestration model’s reasoning effort. Accepted values are minimal, low, medium, high, xhigh, and max. Each model uses a subset of those values, and a request above that subset is lowered to the highest level the model accepts. Omit the field to keep the default for the model.
• instructions.system appends extra system instructions to the managed coding-agent prompt. The text does not replace the sandbox, tool, or skill instructions.
• stage_mounts, on the code_toolset_all entry in tool_resources, mounts a Snowflake stage or versioned stage into the sandbox. Versioned stages include notebooks, Streamlit apps, workspaces, and agents. workspace_mounts is unchanged and still mounts a persistent, writable workspace at version live.

For more information, see Coding Agent.