Skip to main content

ConfirmableActionTool

headless_lms_chatbot::chatbot_tools::action_tools

Trait ConfirmableActionTool 

Source 
pub trait ConfirmableActionTool: ChatbotToolDeclaration {
    type Arguments;
    type Facts;

    // Required methods
    fn parse_arguments(arguments: &str) -> ChatbotResult<Self::Arguments>;
    fn call_requirements(
        arguments: &Self::Arguments,
        user_context: &ChatbotTurnContext,
    ) -> Vec<ToolRequirement>;
    fn execute(
        conn: &mut PgConnection,
        app_config: &ApplicationConfiguration,
        arguments: &Self::Arguments,
        authorization: &ToolAuthorization<Self>,
    ) -> impl Future<Output = ChatbotResult<(ExecutedAction, Self::Facts)>> + Send
       where Self: Sized;
    fn output_description_instructions(
        arguments: &Self::Arguments,
        facts: Option<&Self::Facts>,
        app_config: &ApplicationConfiguration,
    ) -> Option<String>;

    // Provided method
    fn declined_output(_arguments: &Self::Arguments) -> String { ... }
}
Expand description

A client tool whose confirmed answer performs a privileged mutation on the server.

Suspends the turn like ClientChatbotTool; the client answers with ConfirmAnswer. A declined answer produces Self::declined_output without touching the database. A confirmed answer runs Self::execute inside the same transaction as the audit insert and the recorded tool output, guarded by lock_unanswered_for_execution so it can run at most once.

Required Associated Types§

Source

type Arguments

Source

type Facts

Extra facts Self::execute gathers while performing the mutation, used only to decide what Self::output_description_instructions should mention for this particular call. () for a tool with nothing worth gating on.

Required Methods§

Source

fn parse_arguments(arguments: &str) -> ChatbotResult<Self::Arguments>

Source

fn call_requirements( arguments: &Self::Arguments, user_context: &ChatbotTurnContext, ) -> Vec<ToolRequirement>

What the caller must be allowed to do against what this call actually targets. Checked before the turn suspends on the call and again immediately before the mutation runs.

Source

fn execute( conn: &mut PgConnection, app_config: &ApplicationConfiguration, arguments: &Self::Arguments, authorization: &ToolAuthorization<Self>, ) -> impl Future<Output = ChatbotResult<(ExecutedAction, Self::Facts)>> + Send
where Self: Sized,

Performs the mutation. Must re-verify every model-supplied display field against the database (case-insensitively for emails) and refuse with a descriptive error rather than mutate on a mismatch, so a wrong or stale display can never mutate a row it doesn’t describe. authorization both proves the confirming admin was checked against this call’s own requirements and names them, so the mutation cannot run ahead of the check and the audit row cannot credit anyone else.

Source

fn output_description_instructions( arguments: &Self::Arguments, facts: Option<&Self::Facts>, app_config: &ApplicationConfiguration, ) -> Option<String>

Just-in-time instructions for the LLM on how to describe the outcome. facts is None on a declined answer (nothing executed) and Some after a successful Self::execute, so instructions can be conditioned on what this call actually did instead of always including everything that could ever be true. app_config is available on both branches (unlike facts) so a verification link can be offered even on a decline.

Provided Methods§

Source

fn declined_output(_arguments: &Self::Arguments) -> String

Dyn Compatibility§

This trait is not dyn compatible.

In older versions of Rust, dyn compatibility was called "object safety", so this trait is not object safe.

Implementors§