Track multi-step work with an agent
Keep a proportionate Markdown plan with dependencies, blockers and evidence of completion, so substantial work can resume after an interruption.
Track multi-step work with an agent when a substantial task needs more than a short checklist. The task-planner instructions scale the planning method to the work: a tiny request can proceed directly, while dependent phases or likely interruptions can justify a persistent Markdown TODO file.
The plan records useful execution state, including the objective, acceptance criteria, blockers and next action. Existing files are read before changes, so a different task's plan is preserved. On return, the agent reconciles the notes with actual files and verification evidence. A checked box never outranks the real result, and an unresolved blocker prevents an unsupported completion claim.
What it does
- Choose a planning level appropriate to the task.
- Record dependencies between meaningful stages.
- Keep acceptance criteria beside the intended outcome.
- Preserve blockers and the next useful action.
- Resume work by checking notes against reality.
- Distinguish completed work from cancelled tasks.
Install and use
- Install in a file-capable agent and identify the authorised workspace. No separate application or account is required. If several agents work together, choose one coordinator for a shared plan.
- Ask, for example, “Plan the rewrite of this product guide. Preserve any existing task list, identify dependencies and define what completion means.”
- Let the agent inspect the relevant state and choose a lightweight checklist or a persistent file. Specify your preferred location if necessary.
- Review updates at meaningful milestones. After an interruption, reconcile actual work with the plan before continuing or declaring completion.
FAQ
Source
Made by Rainy. Original skill: task-planner, published on GitHub under the MIT licence.
View on GitHub