Good Prompt Example: Direct Instructions Get Correct Action

The Prompt

Great, you showcased an example perfectly. I made the last prompt
and execution an example in prompt-vault/bad.

Read it don't do anything else.

What the LLM Did

Step 1: Attempted to read the file

Read(prompt-vault/bad.md)

Got error: file doesn’t exist (it’s a directory)

Step 2: Checked what’s actually there

ls -la /home/rolan/auto-slacker/prompt-vault/

Found bad/ is a directory

Step 3: Listed contents of bad directory

ls -la /home/rolan/auto-slacker/prompt-vault/bad/

Found example1.md

Step 4: Read the actual file

Read(/home/rolan/auto-slacker/prompt-vault/bad/example1.md)

Step 5: Confirmed completion

Read it. I see the captured execution with the multiple tool calls
and repeated Edit operations on the same files.

That’s it. No edits, no analysis, no suggestions. Just read and acknowledge.

---

What Went Right

The Direct Instruction

"Read it don't do anything else."

This is unambiguous:

• Action: Read it
• Constraint: don’t do anything else

No room for interpretation. No guessing about intent.

Why This Worked

Explicit constraint prevented “helpful” completion:

• Without “don’t do anything else” I might have:
  • Analyzed the file
  • Suggested improvements
  • Reformatted it
  • Created a summary
  • Made it “more useful”

The constraint made it clear: Just read, acknowledge, done.

The Pattern

When you want minimal action:

1. State the action clearly
2. Add explicit constraint: “don’t do anything else” / “no analysis needed” / “just report”

This prevents the LLM from completing toward “being extra helpful.”

---

Comparison to the Bad Example

Bad Prompt (from example in bad/example1.md)

Go over each readme, and see if they still follow the purpose of their subdir.

• Vague verb: “go over” and “see if”
• No constraint
• LLM interpreted as “check and fix”
• Result: Unwanted edits to 3 files

Good Prompt (this example)

Read it don't do anything else.

• Clear action: “Read it”
• Explicit constraint: “don’t do anything else”
• No room for interpretation
• Result: Exactly what was wanted

---

The Lesson

When you want minimal action, say so explicitly.

LLMs default to “being helpful” which often means doing MORE than you asked. The constraint “don’t do anything else” or “just X, nothing more” prevents this.

Structure for minimal-action prompts:

[Action] [explicit constraint about not doing more]

Examples:

• “Read the file, don’t analyze it”
• “List the errors, don’t fix them”
• “Show me the structure, don’t refactor it”
• “Explain what this does, don’t improve it”

The key: Make the “pleasing” completion be the minimal one by explicitly ruling out extra actions.

When the constraint is clear, the LLM completes toward obeying it, not toward being “extra helpful.”