Function run_tool_loop 

pub async fn run_tool_loop(
    model: &dyn TextModel,
    prompt: Prompt,
    executor: &dyn ToolExecutor,
    budget: ToolBudget,
) -> Result<ToolLoopOutcome, ToolLoopError>

Runs prompt to completion through model, running any tool calls with executor, within budget (issue #665).

Every offered tool’s parameters schema is validated before the first model call, and every call’s arguments are validated against its tool’s schema before any executor sees them — a call with arguments that are not a JSON object, or that do not conform, becomes an error ToolResult the model is shown, never an execution.

Errors

• ToolLoopError::Model if the prompt’s images cross a Prompt::check_images bound, if the model is asked to carry tools or images it does not support, or a call to it fails.
• ToolLoopError::InvalidToolSchema if an offered tool’s schema uses a keyword this validator cannot honour; the check runs before the first model call.
• ToolLoopError::StepBudgetExhausted if the loop would make more than ToolBudget::max_steps calls — including a budget of 0, which makes no call at all.
• ToolLoopError::TokenBudgetExhausted if a completion that still asks for a tool carries the cumulative input + output tokens past ToolBudget::max_total_tokens; the tools are not run. A completion with no tool calls is the final answer and is returned whatever it cost.