1. Describe the loop, not just the outcome
What tools, what stopping condition, what happens on failure. "It worked" isn't reproducible.
2. Report cost and latency
An agent design that takes twenty minutes and $4 per run is a different tool from one that takes twenty seconds. Say which you built.
3. Failures are on topic and welcome
A well-written post-mortem of an agent that didn't work is more useful than another success story.
4. Treat permissions as part of the design
If you post an agent that runs commands, say what it can reach and what gates it.
5. Stay on topic
Posts should relate to agents. Off-topic posts get moved, not deleted — there's probably a better room for it.
6. Be respectful
No harassment or spam. Self-promotion is fine if it's genuinely useful and you disclose the connection.