Conclusion

#If you only remember three things

• Brief it well. Three lines (goal, constraints, context) beats a vague paragraph every time. If output is generic, the prompt was probably generic.
• Verify, don't trust. Assistants sound confident even when guessing. For anything that matters — addresses, dates, code, claims about the world — ask them to show their work or check.
• Write things down. Each session resets. If something should persist to tomorrow, save it to memory now.

#What this guide doesn't cover (yet)

On purpose — to keep it short. Topics worth a v0.3:

• Cookbook / copy-pasteable prompts for common situations
• Cost mental model — what a typical agent run costs in real dollars
• Multi-agent patterns — when to fork sub-agents vs run everything in one
• Model comparison — Claude vs GPT vs Gemini, when to switch
• Failure case studies — the stories behind some of the rules

#The honest meta-take

#What to do next

• If you don't have an assistant set up yet: start with Setup & install. ~30 min from zero to a working assistant.
• If you do: open it, ask it to read this guide, and ask which sections most apply to how it's being used. Then act on the gap.
• If you're building assistants for others: the security + limitations pages are non-negotiable. The rest is taste.
• Share this guide with anyone you'd like to be a better user of these tools. Designed to be sent as-is.

#One last thing

This whole site was written by an AI assistant and edited by a human in a single afternoon. That's a small data point about what a well-briefed agent + a present human can do together.

Get the loop right and the work feels different.