Chatgpt Ignores Custom Instructions And Won T Stop 61B988Af

What This Skill Does

This skill addresses a common frustration with LLMs, specifically ChatGPT, regarding the pervasive use of "phantom contrast framing." This speech pattern involves the model creating a non-sequitur strawman argument or a false interpretation of a user's input, only to immediately refute it with a "that's not X; it's Y" structure. This behavior often persists despite explicit user custom instructions to stop. This skill acts as a corrective prompt designed to override the model's tendency to utilize these repetitive and unnatural conversational structures, ensuring that the AI focuses on direct, helpful, and concise responses rather than forced rhetorical refutations.

Installation

To integrate this skill into your environment, use the OpenClaw command-line interface. Run the following command in your terminal:

clawhub install openclaw/skills/skills/hhhh124hhhh/chatgpt-ignores-custom-instructions-and-won-t-stop-61b988af

Ensure that your OpenClaw environment is updated to the latest version to maintain compatibility with prompt-injection correction patterns.

Use Cases

This skill is highly valuable for users who require the AI to act as a direct advisor or technical collaborator. It is particularly effective for:

1. Professional Correspondence: Ensuring the AI does not adopt an argumentative or defensive tone in professional emails.
2. Creative Writing: Removing robotic, formulaic speech patterns that break the immersion of narrative text.
3. Technical Troubleshooting: Reducing the "fluff" and unnecessary hedging when you need clear, actionable data or code.
4. Instructional Design: Getting precise answers to questions without the AI attempting to "reframe" your intent.

Example Prompts

1. "Analyze my project logs. Remember, do not use the 'that's not X; it's Y' framing style; just give me the data."
2. "Explain the fundamentals of quantum computing, and stick to the facts without the phantom contrast strawman structures.
3. "Write a summary of this report. If you attempt to reframe my request using a 'not that, but this' structure, restart your response immediately."

Tips & Limitations

While this skill is highly effective, it operates on a prompt-engineering basis. It does not patch the underlying model's weight or bias directly. If the model is heavily prompted by other system instructions, you may need to re-invoke this prompt to maintain consistency. Users should also note that some specialized versions of models are fine-tuned to prefer a specific "balanced" conversational style, which may occasionally trigger these patterns even with corrective prompts. For best results, chain this skill at the beginning of your chat session to set a persistent guardrail.