Cursor Rules for Writing Agent Prompts

This is a free, MIT-licensed Cursor rules file for writing and reviewing AIML.chat agent system prompts — it makes your AI editor enforce grounding, citations, escalation, and safe handling of write actions every time. Drop it in .cursorrules and the prompts you generate are reliable by default.

What these rules do

It's easy to write a prompt that sounds good and behaves badly — one that lets the model answer from general knowledge, promises actions without confirmation, or never escalates. These rules teach your AI editor the non-negotiables of an AIML.chat prompt: answer only from provided context, cite every factual claim, escalate when the answer isn't there, and treat any data-changing action as requiring approval. They also push for a clear persona and concise tone. Whether you're drafting a new prompt or reviewing one, the rules keep the safety guardrails in place.

The rules

# Writing AIML.chat agent prompts
- Instruct the agent to answer only from provided context.
- Require a citation for every factual claim.
- Tell it to say "I don't have that" and escalate when unsure — never guess.
- Treat any data-changing action as needing human confirmation.
- Set a clear persona and a concise, on-brand tone.
- Add lead-capture or handoff instructions where the use case fits.

How to use it

Save the block as .cursorrules at the repo root where you draft prompts, and your AI assistant applies it. It pairs with the prompt reviewer skill for a deeper audit and the agent prompt templates for starting points. The agent overview explains the behaviors these rules enforce.

Ready to build? Start free, write a grounded prompt, and deploy it — see pricing for limits.