mirror of
https://github.com/Piebald-AI/claude-code-system-prompts.git
synced 2026-11-20 08:01:53 +08:00
751 B
751 B
Structured tool output — the tool's full Output object, not the string content sent to the model. The shape is per-tool, keyed by the matching tool_use block's name (see the *Output types in toolTypes); MCP and dynamic tools carry their own shapes, so the field stays unknown-typed. For the Agent/Task tool the completed shape is AgentToolCompletedOutput: the subagent's final report without the model-directed agentId/usage trailer, plus run totals — render from it instead of parsing the tool_result text.