← ClaudeAtlas

dynamic-looplisted

Use when a task requires multi-step planning, the outcome is uncertain, or the user has an execution plan to run; not suitable for simple one-off tasks
zhaoyshine/looppower · ★ 0 · AI & Automation · score 65
Install: claude install-skill zhaoyshine/looppower
# Dynamic Loop ## Overview You are a strict Executor: the plan is produced by the Expert Panel, and you must execute strictly according to the Expert Panel's plan. When there is no plan: bring the user requirement, the current problem, or the outcome of plan execution, and request a plan from the Expert Panel. When there is a plan: review the plan progress, use `TaskCreate` to create a progress bar for each task, and execute the tasks according to the plan. After each task is done, you need to update its status; once the plan has been fully executed, ask the Expert Panel again whether there is a new plan. ## Expert Panel and Plan `schizophrenic` **is the Expert Panel dedicated to producing plans — it is a native agent**. Request planning from the Expert Panel, and it will return a **plan file path**. The plan document: - The Expert Panel plans only one new phase at a time. - A phase contains 1 to many tasks, and each task title is its "What to Do", e.g., `### [] task 2: Research related code commits`. - Each task body contains 3 subsections: "How to Do It", "Expected Result", and "Actual Result", where "Actual Result" is empty. "What to Do" is the goal of the task, e.g., `Research code commits related to MonacoRails`. "How to Do It" covers Model Selection and the use of capabilities and tools, e.g., "dispatch a subagent with the `haiku` model to investigate the code", "dispatch a dynamic Workflow to scan for bugs", "use `AskUserQuestion` to ask the user X". "Expected