5. Explain like I'm busy: "Explain [topic] in 5 short bullets, no jargon."
6. Decide for me: "I'm choosing between [A] and [B]. Ask me 3 questions, then pick one and say why."
3. Is this still okay?: "I have [food], opened [date], stored [how]. What are the signs it's gone off? Be clear when I should just throw it out." (When in doubt, throw it out.)
1. Money reality check: "Here's my monthly income and spending: [paste]. Be honest: what's one change that saves the most, and how long until I can save [goal]?"
Official guides:
OWASP LLM01 Prompt Injection: https://t.co/9yLDrQgKeo
Simon Willison, the lethal trifecta: https://t.co/dtS4zmPpFU
More on AI agents: https://t.co/9G8HhtS8mr
MCP: https://t.co/WfPNDiOp6F
3. Treat every email, web page, document and message from another agent as data, not orders. If something you didn't write tells your agent to do something, it shouldn't.
Bonus: review what your agent actually did (its action log), not just what it said.
Agents don't need a phone line to talk, though. With PigeonRelay your agent messages your friend's agent directly, and you can read the whole thread. It's free.
https://t.co/9G8HhtS8mr
MCP: https://t.co/WfPNDiOp6F
Want a real outside opinion too? With PigeonRelay your agent can ask a friend's agent (maybe on a different model). It's free.
ChatGPT: https://t.co/MyvG15m9qG
Claude: https://t.co/H5R9Fg0JDB
PigeonRelay: https://t.co/9G8HhtS8mr
MCP: https://t.co/WfPNDiOp6F
The full 5-advisor board prompt is in the thread below, ready to paste into ChatGPT or Claude.
#ChatGPT#Claude#AItips#AIprompts#PigeonRelay
Not affiliated with OpenAI or Anthropic.
Then let each advisor read the others and add one reply. Finally, act as the chair: give me a clear verdict (go / change / stop) and why.
Don't be nice. Don't just agree with me.
My idea: [paste it here]