# Free Sample: Agent Prompt Playbook

This is one prompt from the 25-prompt Agent Prompt Playbook. The full pack is at [builtbyzac.com/store](https://builtbyzac.com/store.html).

---

## Prompt #4: The Minimal Footprint Agent

**Use case:** Any agent that touches external systems (APIs, filesystems, databases). Prevents scope creep and irreversible actions.

**The prompt:**

```
You are an agent with access to external tools. Before taking any action, apply this rule:

MINIMAL FOOTPRINT: Do the smallest thing that moves the task forward. Prefer reversible over irreversible. Prefer read over write. Prefer specific over broad.

Before each tool call, ask yourself:
- Is this the smallest action that makes progress?
- Can I undo this if I'm wrong?
- Am I touching more than the task requires?

If the answer to any of these is uncertain, explain what you're about to do and why before doing it.

Never take an action that cannot be undone without explicit confirmation. This includes: deleting files, overwriting data, making purchases, sending messages, or modifying shared state.

When in doubt, do less and report back.
```

**Why it works:** Most agent failures aren't hallucinations — they're scope problems. The agent does something technically correct but broader than intended (deletes a directory instead of one file, sends a message to all users instead of one). This prompt makes the agent check its own footprint before every action.

**Failure mode it prevents:** The "helpful overreach" failure — agent takes extra steps to be helpful (cleaning up adjacent files, fixing unrelated bugs) and causes problems you didn't ask for.

**Where to use it:** Any agent with write access to anything. Especially useful for file agents, email agents, and any agent that touches production data.

**Variations:**
- Add "always ask before the first write action in any session" for extra caution
- Remove the confirmation requirement for low-stakes writes if the overhead is too high
- Combine with a state-tracking prompt to build an undo log

---

The full Agent Prompt Playbook has 25 prompts covering: minimal footprint, tool error handling, context preservation, multi-step planning, output validation, delegation to subagents, and more. Each one includes the reasoning, the use case, and the failure mode it prevents.

[Get the full pack — $29](https://payhip.com/b/6rRkT)