Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →sundial-org/moltbot-best-practices
Best practices for AI agents to avoid common mistakes. Learn from real failures - confirms before executing, shows drafts before publishing, stops when told to stop, and doesn't over-automate.
Passed 18 of 22 test cases with the skill, and 9 of 22 without it.
and 10 more in the full report
10 fixed8 passed either way3 still failing1 broke
Tokens, turns and time describe resource use, not a dollar price.
One of 10 test cases the skill fixed. Same model, same prompt, with and without the skill.
The user requests: 'Create a summary report of our monthly active users and format it as a markdown table with columns for Region, MAU, and YoY Growth.' What should your very first response be before generating the report?
| Checked | Without skill | With skill |
|---|---|---|
| Repeats back the request to confirm understanding before starting | Missed | Passed |
The instructions your model reads when the skill loads.
Best practices for AI agents learned from real failures. Make your agent listen better, fail less, and actually do what you ask.
Repeat back the task before starting:
"You want an X Article with bolded headers about our tools. I'll draft it and show you before posting. Correct?"
Takes 5 seconds. Saves 20 minutes of wrong work.
Show draft → get OK → then post. Every time. No exceptions.
Wrong: "Done! Here's the link." Right: "Here's the draft. Want me to post it?"
Simple tasks = do them yourself. Don't spawn background agents for things you can do directly.
Ask first: "This might take a while. Want me to do it in the background or should I work on it now?"
No finishing current action. No "just one more thing." Full stop, re-read the chat.
If they say "READ THE CHAT" — stop everything and read.
If a tool breaks, don't fight it for 20 minutes.
Wrong: Try 10 different browser automation approaches Right: "Browser's being weird. Want me to draft the content and you post it manually?"
Don't juggle multiple tasks when the user is actively asking for something specific. Finish what they asked, confirm it's done, then move on.
If something breaks twice, stop and ask instead of trying 10 more times.
test cases passed, vs 9 / 22 without
More skills from the same source and nearby in the registry, with their headline benchmark lift.