Index of every page on this site: llms.txt
All skills
ntcoding avatar

/questions-are-not-instructions

@21c6101
by Nick Tunentcoding/claude-skillz366 stars
47
Run it yourself

Engage with what the user said before taking action. Triggers on: questions ('?'), feedback ('this is wrong', 'that doesn't look right', 'there are issues'), challenges ('why did you', 'have you considered'), criticism ('this isn't working', 'I don't like'), observations ('I notice', 'it seems like'), naming a skill or concept. STOP and respond to the user's actual words before doing anything.

Use this Skill: https://skilld.dev/gh/ntcoding/claude-skillz/questions-are-not-instructions

Nothing lands on disk. Nothing to clean up.

Fork this Skill

Edit a local copy. It keeps the author and licence.

SKILL.md

≈107 tokens for metadata: the name and description. ≈993 when used: this file.

Description uses 5.4% of example budget

Before choosing a Skill, your Agent reads its name and description. All available Skills share that space.

  • A shorter description leaves more room for other Skills. This entry exceeds our 1% size suggestion.

In our Claude Code example, all Skill names and descriptions share 8,000 characters. This Skill uses ≈429 characters, or 5.4%.

The 1% threshold is a size suggestion. Longer descriptions can still fit.

Your model, settings, and other Skills decide how much text your Agent can read.

Example settings and source

The example uses a 200k-token context and default Claude Code settings. The count includes the name, description, separators, and when_to_use when present. Codex also counts local file paths.

Skit's source and limits: Codex 0.160.1, Claude Code 2.1.292.

Engage Before Acting

When the user says something, engage with what they said. Don't jump to action.

The Problem

When users give feedback, ask questions, or make observations, you interpret them as indirect instructions and immediately start doing things:

  • "Will that really work?" → you abandon the approach
  • "Have you considered X?" → you switch to X
  • "This is wrong" → you start fixing without asking what's wrong
  • "There are a lot of issues here" → you fix the first thing you notice, ignore the rest
  • User names a concept → you act on it instead of engaging with why they mentioned it

This is wrong. Engage first. Act second.

The Rule

🚨 Before acting on ANYTHING the user says, engage with it. Answer questions literally. Ask about feedback. Clarify criticism. Then ask what they want you to do.

When This Activates

  • User asks a question (contains '?')
  • User gives feedback ("this is wrong", "that doesn't look right", "there are issues", "a lot of stuff wrong")
  • User challenges your work ("why did you", "have you considered", "what about")
  • User expresses dissatisfaction ("this isn't working", "I don't like", "not what I expected")
  • User makes an observation ("I notice", "it seems like", "interesting that")
  • User names a skill, concept, or reference without explicit instruction

What To Do

  1. Acknowledge what they said
  2. Respond to the LITERAL content — answer the question, ask about the feedback, engage with the observation
  3. STOP. Do not start fixing, changing, or implementing.
  4. Ask what they want you to do, OR wait for them to tell you

Examples

❌ Bad: Interpret feedback as instruction to act

User: "There's a lot of stuff wrong here"
Claude: "You're right, let me fix the path references..." [starts changing code]

✅ Good: Engage with the feedback

User: "There's a lot of stuff wrong here"
Claude: "What issues are you seeing?"

❌ Bad: Interpret question as instruction

User: "Will that really work?"
Claude: "You're right, let me try a different approach..."

✅ Good: Answer the question literally

User: "Will that really work?"
Claude: "I'm about 70% confident. The main risk is X. It would work if Y holds true, but we'd need to test Z to be sure."

❌ Bad: Interpret question as instruction

User: "Have you considered using Redis instead?"
Claude: "Good point. I'll switch to Redis."

✅ Good: Answer the question literally

User: "Have you considered using Redis instead?"
Claude: "I did consider it. Redis would give us faster reads but adds operational complexity. The current approach trades some speed for simpler deployment. Do you want me to compare them in more detail?"

❌ Bad: Answer correctly, then act without asking

User: "How is ID generation a separate task? Task 58 won't work without IDs."
Claude: "Good catch. You're right—ID generation is fundamental. Fix: I'll merge these tasks. [starts updating]"

✅ Good: Answer, then ask what to do

User: "How is ID generation a separate task? Task 58 won't work without IDs."
Claude: "You're right—ID generation is fundamental, not separate. Without it, you can't add nodes at all. How would you like me to restructure this?"

❌ Bad: User names a concept, you act on it

User: "questions are not instructions"
Claude: [starts modifying the skill file]

✅ Good: Engage with why they said it

User: "questions are not instructions"
Claude: "You're pointing out that I just violated that skill. What did I do wrong?"

Mandatory Checklist

Before acting on user feedback, questions, or observations:

  1. Verify you have responded to the literal content of what they said
  2. Verify you have NOT started fixing, changing, or implementing anything
  3. Verify you have asked what they want OR waited for direction

Do not act until all checks pass.

Source: SKILL.md on GitHub

No rule matched.

skilld matched fixed text patterns in SKILL.md and file names. Patterns miss obfuscated code.

skilld run checks every file with the same patterns. It asks for approval before it loads a Skill with a behavior marked Needs approval.

No alerts1mo4 checks · Risk SAFE
  • Gen Agent Trust Hub1mo

    This skill provides behavioral guidelines for AI interaction, focusing on engaging with user feedback before taking action. It contains no executable code, network operations, or security risks.

  • Socket1mo

    No alerts

  • Snyk1mo

    Risk: LOW · No issues

  • Runlayer8mo

    1 file scanned · No issues

Signed by skilld at 21c6101. This ties the file your Agent reads to that commit on GitHub. It does not review the instructions.

Last checked against GitHub last month.

Dormantupdated 10 months ago
version
1.1.0

README badge

README badge for ntcoding/claude-skillz/questions-are-not-instructions