Struct LlmPlanner 

pub struct LlmPlanner;

LLM-powered planner that generates plans, extracts goals, and evaluates achievement

Implementations

impl LlmPlanner

pub async fn create_plan( llm: &Arc<dyn LlmClient>, prompt: &str, ) -> Result<ExecutionPlan>

Generate an execution plan from a prompt using LLM

pub async fn extract_goal( llm: &Arc<dyn LlmClient>, prompt: &str, ) -> Result<AgentGoal>

Extract a goal with success criteria from a prompt using LLM

pub async fn check_achievement( llm: &Arc<dyn LlmClient>, goal: &AgentGoal, current_state: &str, ) -> Result<AchievementResult>

Evaluate whether a goal has been achieved given current state

pub fn fallback_plan(prompt: &str) -> ExecutionPlan

Create a fallback plan using heuristic logic (no LLM required)

pub fn fallback_goal(prompt: &str) -> AgentGoal

Create a fallback goal using heuristic logic (no LLM required)

pub fn fallback_check_achievement( goal: &AgentGoal, current_state: &str, ) -> AchievementResult

Create a fail-closed fallback achievement result (no LLM required).

Goal completion is a control-plane decision: a host may stop an unbounded engineering loop as soon as GoalAchieved is emitted. Plain answer text such as “done” (or even “not complete”) is not sufficient evidence for that decision. If the structured evaluator is unavailable, keep every success criterion open and let the host retry in a later iteration.

pub async fn pre_analyze( llm: &Arc<dyn LlmClient>, prompt: &str, ) -> Result<PreAnalysis>

Perform pre-analysis in a single LLM call: intent classification, goal extraction, execution plan, and input optimization. Falls back to heuristics on failure.