Runner

---

Object Hierarchy:

Description:

Content:

Properties:

• public List completed { get; }
  Tasks that have gone through execution (unchanged reference data).
• public bool in_replay { get; set; }
  True when running from a replay session. When set, Details and Tool use the runner's chat_call (ReplayChat) instead of their own.
• public List pending { get; set; }
  Tasks to be run (initial plan or new/revised from iteration). Only this list is ever run.
• public ProgressList progress { get; }
  Task progress list model for the 7.14 UI; OLLMcoder.Task.ProgressList holds a weak ref to this runner.
• public bool replay_as_new { get; set; }
  When false (default): hydrate task graph from the transcript; rely on existing task_dir files on disk. When true: also mirror task-list markdown writes during replay (behaviour TBC — see on_replay sketch agent-issues placeholders).
• public Definition skill { get; }
• public Factory sr_factory { get; }
• public Document? user_request { get; set; }
  Used by task list flow (send_async); set from template user_to_document().
• public bool writer_approval { get; set; }
  True once user has approved running writer (modify) tasks this run.

Creation methods:

• public Runner (Factory factory, SessionBase session)

Methods:

• public string env ()
  Used only in send_async when filling task_creation_initial (before user_request exists).
• public PromptTemplate iteration_prompt (string previous_proposal_issues, List existing_proposed, string previous_proposed_md) throws Error
  Build the task list iteration prompt (task_list_iteration.md). Uses completed and existing_proposed (outstanding) and optional previous_proposed_md when retrying.
• public override void on_replay (Message m)
  Apply one stored transcript message during GTK session restore. Dispatches on replay_phase and message role; uses ResultParser on content-stream where live code would have used the LLM response string.
• public async void replay (ArrayList<Message> messages)
  Replay the task-list flow using the session's raw message list. Swaps in ReplayChat and calls send_async so the same code path runs (task creation retry loop, parse, handle_task_list). ReplayChat returns the next session content message on each send().
• public async void run_task_list_iteration (Cancellable? cancellable = null) throws Error
  Task list iteration: send current list to LLM, parse response into this.pending. On parse/validation failure restores this.pending = existing_proposed; uses raw LLM response as previous_proposed for next retry. On failure after 5 tries, cancels the given cancellable so the whole request stops (no way to carry on).
• public override async void send_async (Message in_message, Cancellable? cancellable = null) throws Error
  Entry point. Sends user request only; when finished calls handle_task_list. Current file and open files come from this.project_manager.
• public PromptTemplate task_creation_prompt (string user_prompt, string previous_proposal, string previous_proposal_issues, Manager skill_catalog, ProjectManager project_manager) throws Error
  Build the task creation prompt (task_creation_initial.md). Used by send_async before user_request exists.

Signals:

• public signal void replay_step (string step, string content)
  Emitted during replay before each logical step. Arguments: step name and content (e.g. raw text being parsed; empty string when none). Listeners can show content and wait for user (e.g. Enter) before returning.

Inherited Members: