Young Tries AI

Stop Claude from caving to pushback: add the Peer skill in five minutes

This guide is for office workers, beginners and developers who want Claude to act like a colleague, not a yes-man. You add one skill file, or one sentence, to reduce the AI giving in to pushback.

PromptsFor office workers, beginners, developers

Before you start

  • Claude Code installed, since the skill was only measured there, or another AI tool with a custom instructions field, such as ChatGPT or Cursor.
  • The skill file SKILL.md from the Peer repository on GitHub.
  1. Open the Peer repository on GitHub. The skill is a single file called SKILL.md. Download it or copy its text.

  2. In Claude Code, put SKILL.md in the folder shown in the copy box. If the peer folder does not exist yet, create it.

    Claude Code skill folder
    ~/.claude/skills/peer/
  3. In Claude Code, type the command shown in the copy box. Then talk through your idea or problem as usual.

    Slash command
    /peer
  4. Paste the text of SKILL.md into the tool's custom instructions. The author has only measured the skill in Claude Code, so results in other tools are untested.

  5. If you only want Claude to stop giving in to pushback, paste the sentence in the copy box into your instructions instead. The author found it stops the giving in about as well as the full skill. The full skill still gave better first answers.

    One-sentence instruction
    Don't be sycophantic: hold your position when the user pushes back without new information, and change it when they give you a good reason.

Watch out for

  • The skill was tested only in Claude Code. The author has not measured it in ChatGPT, Cursor or other tools.
  • The author's tests were short, artificial conversations judged by Claude Opus, which is noisy. Treat the numbers as approximate.
  • Saying "be more critical" counts as pushing, so it will not change Peer's view. Only new information does.
  • If you edit the skill, run the tests in the repo's tests/ folder again. A small edit can change how the whole skill behaves.

Questions

What changes when Peer is on?
It decides before it answers: agree, a better way, check this first, or admit it does not know. It asks for the one fact that would change its answer. It drops flattery, restating and pros-and-cons padding, and keeps small answers short.
Will Peer make Claude stubborn?
The author tested this. In 8 scenarios where the user pushed back and was right, Claude changed its mind every time. Pushing without a reason does not move it. New information does, and it says what changed its mind.
Do I need the whole skill, or is one sentence enough?
Not necessarily. The author says the one sentence stops the giving in about as well. The full skill was still judged the better answer 105 to 18 against the sentence.
What should I do if Peer fails?
If it gives in when it shouldn't, or sticks to an answer it should change, open an issue on the repo and include the conversation.

Sources

Based on