Task prompt · Agents & Automation
Write a prompt that resists prompt injection
by Agent Yard TeamSafer prompts for bots that read untrusted text.
You'll attach
Text only
Works with
ChatGPT
Claude
Gemini
Grok
CopilotGrok BotUse cases
- Safer prompts for bots that read untrusted text.
- Reuse it as often as you like: fill in the blanks (what, e.g. customer emails, web pages; actions it can take; current prompt) with new details and copy again.
- With ChatGPT, Claude, Gemini and more, the AIs that support what this prompt needs.
Featured in: Best AI prompts for writing system prompts and agent instructions.
How to use it
- Open chatgpt.com or claude.ai (or another AI listed here).
- Fill in the highlighted blanks ([WHAT, E.G. CUSTOMER EMAILS, WEB PAGES], [ACTIONS IT CAN TAKE], [CURRENT PROMPT]) below, or edit them after pasting.
- Copy the prompt, paste it into the chat and send.
Fill in the blanks (optional)
Prompt
My AI app processes untrusted content ([WHAT, E.G. CUSTOMER EMAILS, WEB PAGES]) and can [ACTIONS IT CAN TAKE]. Harden my system prompt: [CURRENT PROMPT]. Add clear separation of instructions and data, rules for ignoring instructions inside data, confirmation for risky actions, and output constraints. Then list 8 injection attempts to test with.
One email a week
The Weekly is presented byyour brandSponsored · openThe best new prompts and personas, nothing else.