regular-deferred-task-runner

name	regular_deferred_task_runner
description	Wrapper skill: each hour, gather pending deferred tasks into the runner todo list and process them one by one.

Regular Deferred Task Runner

This is a wrapper skill for the repo automation in 90_System/Skills/deferred_task_queue/.

Use

• Start by inspecting every task currently in pending
• Add each pending task to your working todo list before execution
• For each task in your working todo list, first perform planning on how to complete it the most efficient way, and only then start the execution.
• Use this as a general deferred-work runner, regardless of which system created the task

Workflow

• List all pending tasks and turn them into explicit todo items for the current run
• Claim the oldest pending task
• Render the deterministic prompt for the claimed task
• Complete the requested work by first planning the most efficient way to do it, and then executing the plan
• Mark the task done with a short outcome summary, or mark it failed and retryable if it should return to pending
• Repeat until no pending tasks remain or the run must stop

Run (from the vault repo root)

• List pending tasks:
  • .\.venv\Scripts\python.exe 90_System\Skills\deferred_task_queue\task_queue.py list --status pending
• Claim the oldest pending task:
  • .\.venv\Scripts\python.exe 90_System\Skills\deferred_task_queue\task_queue.py claim-next --worker desktop_codex
• Render the runner prompt:
  • .\.venv\Scripts\python.exe 90_System\Skills\deferred_task_queue\task_queue.py render-prompt --task 90_System\TaskQueue\running\<task-file>.md
• Complete a task:
  • .\.venv\Scripts\python.exe 90_System\Skills\deferred_task_queue\task_queue.py complete --task 90_System\TaskQueue\running\<task-file>.md --summary "Finished summary"
• Fail a task:
  • .\.venv\Scripts\python.exe 90_System\Skills\deferred_task_queue\task_queue.py fail --task 90_System\TaskQueue\running\<task-file>.md --summary "Failure summary" --retryable

Notes

• This is a regular scheduled workflow, not an ad-hoc queue creation flow.
• Keep summaries short and operational.
• Prefer retryable failure only when the blocker is likely temporary.
• A task may originate from Telegram bridge, another automation, or a manual enqueue request; the runner should treat them uniformly.