Struct TaskPlanner 

pub struct TaskPlanner {
    pub task_list: Vec<Task>,
    /* private fields */
}

Task planner: generates and tracks task lists for the agentic loop.

Fields

task_list: Vec<Task>

Current task list.

Implementations

impl TaskPlanner

pub fn new( workspace: Option<&Path>, chat_root: Option<&Path>, availability: Option<ToolAvailabilityView>, ) -> Self

Create a new TaskPlanner.

workspace: per-project override directory. chat_root: ~/.skilllite/chat/ for loading prompts from the seed system.

pub async fn generate_task_list( &mut self, client: &LlmClient, model: &str, user_message: &str, skills: &[LoadedSkill], conversation_context: Option<&str>, goal_boundaries: Option<&GoalBoundaries>, soul: Option<&Soul>, ) -> Result<Vec<Task>>

Generate task list from user message using LLM.

goal_boundaries: Optional extracted boundaries (scope, exclusions, completion conditions) to inject into planning. Used in run mode for long-running tasks. soul: Optional SOUL identity document; when present, Scope & Boundaries are injected (A8).

pub fn matched_rule_ids(&self) -> &[String]

Return planning rule IDs matched for the current user request.

pub fn sanitize_and_enhance_tasks( &self, tasks: &mut Vec<Task>, skills: &[LoadedSkill], )

Apply sanitize_task_hints and auto_enhance_tasks to a task list (e.g. from replan). Use this when accepting a new plan from update_task_plan so replan has the same validation and enhancement as initial planning.

pub fn build_execution_prompt(&self, skills: &[LoadedSkill]) -> String

Build the main execution system prompt from the external template. Placeholders: {{TODAY}}, {{YESTERDAY}}, {{SKILLS_LIST}}, {{OUTPUT_DIR}}.

pub fn build_task_system_prompt( &self, skills: &[LoadedSkill], goal_boundaries: Option<&GoalBoundaries>, ) -> String

Build system prompt with task list and execution guidance.

goal_boundaries: Optional extracted boundaries to inject (A5, run mode).

pub fn mark_completed(&mut self, task_id: u32) -> bool

Mark a task as completed and return whether it was found.

pub fn all_completed(&self) -> bool

Check if all tasks are completed.

Returns false when the task list is empty: an empty plan means the LLM decided no explicit tasks were needed, which is not the same as having finished a set of tasks. Treating an empty list as “all done” causes run_with_task_planning to fire the “final summary” branch immediately after the first batch of tool calls, ending the loop prematurely.

pub fn is_empty(&self) -> bool

Check if the task list is empty (LLM decided no tools needed).

pub fn current_task(&self) -> Option<&Task>

Get the current (first uncompleted) task.

pub fn build_nudge_message(&self) -> Option<String>

Build a nudge message to push the LLM to continue working on tasks.

pub fn build_depth_limit_message(&self, max_calls: usize) -> String

Build a per-task depth limit message. Suggests complete_task first; if the approach is wrong, suggests update_task_plan.