tool-call-flow

What problem does it solve?

Tool-calling systems often fail in subtle ways because the model, runtime, message history, and tool-result pairing must follow a precise cycle; this Skill explains that protocol so teams can implement and audit it correctly.

Core Features & Use Cases

• Defines the four-phase tool-call cycle: declaration, request, execution, continuation.
• Clarifies message-history as the source of truth for how tool calls and results are represented across turns.
• Compares vendor protocol encodings (Anthropic tool-use, OpenAI function-calling, MCP, Gemini function calling) while preserving the same underlying cycle.
• Covers parallel tool calls and streaming tool calls with correctness implications and dependency limits.
• Specifies failure encoding so errors remain tool-result messages that the model can read and recover from.
• Distinguishes model intent from runtime execution to maintain auditability, composability, and recoverability.

Quick Start

Ask an AI runtime engineer: “Explain the exact four-phase tool-call cycle and show how a tool-result must be appended to the message history before the next model turn, including how failures are encoded.”