How to use the Task Planning agent skill (task-planning)

What the task-planning skill makes the agent do

The skill tells the agent to plan only as much as the work needs. If execution is authorized and the next step is small and clear, the agent should just do it. A request for a plan alone does not require child tasks or a new approval gate.

A plan written under this skill describes the intended outcome, relevant constraints, the chosen approach, and how success will be verified. It also lists uncertainties or decisions that affect execution. The skill asks the agent to keep implementation steps within one owner's task unless another owner, useful parallel output, a dependency, or independent review justifies separate work.

• States the intended outcome and constraints
• Explains the chosen approach and how success will be checked
• Lists open uncertainties or decisions that affect execution
• Saves the plan as the issue document with key plan, updating the current revision rather than duplicating it

How to install the task-planning skill

Download the ZIP for the skill from this page. The skill is a single file, SKILL.md, so the unzipped folder contains that file.

For Claude Code, unzip the skill into ~/.claude/skills/ to make it available for your user account, or into a project's .claude/skills/ folder to make it available only in that project.

In the Claude apps, you can upload the ZIP instead. Go to Settings, then Capabilities, then Skills, and upload the file there.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside the project folder
• Claude apps: Settings → Capabilities → Skills

How to trigger the skill with natural requests

You do not need a special command. The skill's description says to use it when a plan is requested or when uncertainty needs resolving before execution, so ask for a plan in ordinary language. Name the issue or task the agent should work on, and say what format you want if you have a preference, since the skill tells the agent to use your preferred format.

Example requests that should match the skill: asking the agent to plan the work on a Paperclip issue before starting, or asking it to revise an existing plan after new information arrives.

• Write a plan for this issue covering the outcome, approach and how we will verify it.
• Revise the current plan on this task now that the API constraint has changed.
• Before you start, lay out the uncertainties and the decisions that affect execution.

How approvals and planning mode work

The skill tells the agent to respect planning mode and explicit approval requests. When approval is required, the agent asks for confirmation of the latest plan revision and waits for that decision. After acceptance, it executes the cohesive work on the current ordinary task.

If delegation to child tasks is justified, the agent uses the paperclip-converting-plans-to-tasks skill. That referenced skill is needed only in that case.

• Approval required: the agent requests sign-off on the latest revision and waits
• Approval accepted: the agent carries out the work on the current task
• Delegation justified: the agent switches to paperclip-converting-plans-to-tasks

What the skill needs and safety notes

The skill relies on Paperclip's issue, document and planning tools. Agents with native document tools use them directly. Legacy agents follow the mechanics in the paperclip skill's document and planning API. Without Paperclip, the skill has nothing to act on, because the file is only a text instruction.

The skill does not run scripts and does not use the network, according to its listing. Because it is plain instructions, it does not install software or fetch external resources on its own.

• Requires Paperclip with native document tools, or the paperclip skill's document and planning API for legacy agents
• Runs scripts: no
• Uses network: no
• Optional referenced skill: paperclip-converting-plans-to-tasks, only when delegation is justified

Tips for getting useful plans

Ask for a plan when the work is genuinely uncertain or multi-part. For a small, clear next step, tell the agent to proceed, since the skill says not to over-plan. Keep the plan as one revised document on the issue so reviewers always see the current version.

When approval is in place, review the latest revision before you accept it, because the agent will wait for that decision and then continue.

• Skip planning for simple one-off questions
• Ask the agent to update the existing plan rather than start a new one
• Check that the plan includes a verification step