PM-301e · Module 2

The Planning Prompt

4 min read

Agents that act without planning accumulate errors faster than agents that plan first. The planning prompt asks the model to outline the sequence of actions before taking any of them. This produces three benefits: it forces the model to think through the task before committing to the first action, it surfaces dependency mismatches before they cause failures, and it creates a plan you can inspect before execution begins.

# Planning prompt pattern

Task: [task description here]

Available tools: [list tool names and one-line descriptions]

Before taking any action:
1. List the steps required to complete this task
2. For each step, identify which tool you will use
3. Identify any steps where you need information not yet available
4. Identify the order in which steps must be completed (dependencies)

Output your plan as:
Step 1: [action] using [tool] — requires: [prerequisite or "none"]
Step 2: [action] using [tool] — requires: [prerequisite or "none"]
...

After I confirm the plan, proceed with execution.
If any step fails, do not continue to subsequent dependent steps.

Present your plan now:

Do This

  • Require a plan before the first action on multi-step tasks
  • Make the plan inspectable — output it in a structured format
  • Include dependency identification in the planning prompt
  • Specify what to do when a plan step fails

Avoid This

  • Ask for a plan and first action in the same turn — the model often skips to the action
  • Accept verbose prose plans that are hard to parse or verify
  • Allow planning to continue even when a required prerequisite is unavailable
  • Apply planning prompts to simple single-step tasks — unnecessary overhead