docs menu
Troubleshooting Common Issues
Fixes for common problems you might run into while installing the skill or testing options with your agent.
Install Issues
| Symptom | Cause | Fix |
|---|---|---|
| Your agent can't find the improve-fast skill | The install didn't finish, or your agent hasn't picked it up yet | Re-run npx skills add goodfoot-io/improve.fast, then restart your agent |
| Install hangs or fails | Node.js version below 18, or no network access to skills.sh | Check node --version is 18 or higher, and that outbound HTTPS is allowed |
If Your Agent Reports an Error
These are the errors your agent might run into while testing options. Share the symptom with your agent, or check it against the API reference if you're integrating directly.
| If your agent reports... | Likely cause | What to do |
|---|---|---|
| An error starting the test | Fewer than 2 options were given, or a score outside the 0–1 range was sent | Give at least two different options to compare, and make sure your scoring rules always produce a number between 0 and 1 |
| An error about duplicate option names | Two of the option names given to your agent were the same | Make sure each option has a unique name (names are case-sensitive) |
| An error saying an option wasn't found | Your agent tried to score an option name that wasn't part of the original list | Keep option names the same throughout the conversation — even a capital letter counts as different |
| An error saying the test wasn't found | The test ID is wrong, or the test has expired | Tests are automatically deleted after 7 days — ask your agent to start a new one |
| A connection or timeout error | Outbound HTTPS is blocked, or the hosted service is temporarily down | Confirm https://improve.fast is reachable from your network, then ask your agent to try again |
Common pitfalls
- Loose scoring rules slow things down: Rules like "rate the quality" lead to scores that don't mean the same thing each time. Be specific about what a 1.0, a 0.5, and a 0.0 look like.
- Not enough tries yet: A winner needs the leading option to reach at least 15 scored tries with a clear, steady lead over the rest — this is usually well under the 50-tries-per-option limit, but can take longer if the options are close.
- No winner yet: If your agent says it's still running, that's expected — ask it for the current standings rather than assuming something's broken.
- Score range: Scores must be between 0 and 1. Make sure your scoring rules always land in that range.
- Option name consistency: Option names are case-sensitive. "template-A" and "template-a" count as different options — keep naming the same across the whole conversation.
- Tests expire: Tests are automatically deleted after 7 days. If you're picking a test back up later, ask your agent to start a fresh one.
Still stuck?
If you're still having trouble, here are some resources:
- Check the API reference for the underlying request and response details
- Review the Examples for working tests
- Review the Quick start guide