compress-prompt

Compressing a Prompt (opt-in, confirm-first)

This skill shrinks a verbose prompt the user provides. It NEVER runs automatically and NEVER changes meaning. It does not classify or "detect" meaning change (that would require an extra model call — out of scope).

Steps

1. Take the user's draft prompt (the text they want compressed).
2. Produce a compressed version that preserves EVERY requirement, constraint, and piece of intent. Only remove: filler, politeness padding, redundancy, repeated context. Do not drop any instruction, file name, number, or acceptance criterion.
3. Estimate tokens before and after using ~4 characters per token (rough estimate).
4. Show the user:
   • the compressed version in a code block,
   • a short bullet list of what was cut (categories, not a full word-diff),
   • Before: ~N tok -> After: ~M tok (~X% smaller, rough estimate).
5. Ask the user to confirm: "Use the compressed version, or keep the original?"
6. Do nothing further until the user responds. The compressed text is only used if the user re-submits / approves it. If they decline, keep the original unchanged.

Hard rules

• Never auto-apply the compressed prompt.
• Never alter meaning to make it shorter. If a section cannot be shortened without risking meaning, leave it as-is and say so.