Anti Prompt Bleed

An agent skill for preventing instruction and prompt bleed in generated website, app, and frontend UI copy.

This skill is narrowly scoped: it removes user-visible leakage of prompts, hidden instructions, implementation requirements, tool/runtime details, planning residue, and assistant-process wording. It does not rewrite normal marketing copy, brand voice, slogans, or generic phrasing unless that text exposes internal instructions.

Install

Install with the skills.sh CLI:

npx skills add eimexdev/anti-prompt-bleed

Then restart your agent so the skill is picked up.

Use

Invoke it explicitly when reviewing or building frontend work:

Use $anti-prompt-bleed to review this generated website and remove only instruction, prompt, assistant, or implementation leakage from visible UI copy.

What It Catches

• Prompt references such as "As requested" or "Based on your request"
• Echoed design or implementation instructions such as "Use lucide icons" or "Keep cards at 8px radius"
• Tool/runtime mentions such as skills, system prompts, developer messages, commands, local files, or generated assets
• Planning residue such as TODOs, acceptance criteria, implementation notes, or test plans
• UI copy that describes the generation process instead of the product

What It Leaves Alone

• Marketing language, even if generic
• Brand voice, slogans, and section labels
• Phrases like "we built", "designed for", or "crafted for" when they read as normal product copy
• Placeholder or mock content unless it leaks task or prompt wording