Module tool_simulation 

Text-based tool simulation for CLI runners without native function calling

Text-Based Tool Simulation

CLI LLM runners (Claude Code, Copilot, Cursor Agent, OpenCode) communicate via plain text and do not support native function calling. This module provides a text-based tool simulation layer that enables tool calling by:

1. Generating a markdown tool catalog from function declarations and injecting it into the system prompt
2. Parsing <tool_call> XML blocks from LLM text output
3. Formatting tool results as <tool_result> XML blocks for re-injection
4. Running a full multi-turn tool loop that iterates until the LLM produces a final text response

This is the CLI counterpart to the SDK-managed tool calling in CopilotHeadlessRunner (requires copilot-headless feature).

Quick Start

use embacle::tool_simulation::*;
use embacle::types::{ChatMessage, ChatRequest, LlmProvider};
use serde_json::json;
use std::sync::Arc;

let declarations = vec![
    FunctionDeclaration {
        name: "get_weather".into(),
        description: "Get weather for a city".into(),
        parameters: Some(json!({"type": "object", "properties": {"city": {"type": "string"}}})),
    },
];

let handler: TextToolHandler = Arc::new(|name, args| {
    FunctionResponse {
        name: name.to_owned(),
        response: json!({"temperature": 72}),
    }
});

let mut messages = vec![
    ChatMessage::system("You are a helpful assistant."),
    ChatMessage::user("What's the weather in Paris?"),
];

let result = execute_with_text_tools(
    provider, &mut messages, &declarations, handler, 5,
).await?;
println!("{}", result.content);

Structs

FunctionCall

A parsed tool call extracted from LLM text output.

FunctionResponse

A tool execution result to feed back to the LLM.

TextToolResponse

Result of a text-based tool-calling conversation.

Functions

execute_with_text_tools

Execute a full text-based tool-calling conversation with a CLI provider.

format_tool_results_as_text

Format function responses as text for injection into follow-up messages.

generate_tool_catalog

Generate a text-based tool catalog from function declarations.

inject_tool_catalog

Inject a tool catalog into the system prompt of a message list.

parse_tool_call_blocks

Parse <tool_call> blocks from LLM text output into structured function calls.

strip_simulation_artifacts

Strip all simulation scaffolding from a model’s user-facing reply: both <tool_call> blocks (via strip_tool_call_blocks) and echoed tool-result scaffolding (via strip_tool_result_echo).

strip_tool_call_blocks

Strip <tool_call>...</tool_call> blocks from text, returning remaining content.

strip_tool_result_echo

Strip echoed tool-result scaffolding from a model’s reply.

Type Aliases

FunctionDeclaration

A tool definition describing a callable function.

TextToolHandler

Callback type for executing tool calls.