Skip to main content

​
Troubleshooting

​
AI over-promises or fix loops

Symptoms：
  • AI says it will definitely fix it, but the bug remains or new bugs appear
  • Same bug fixed many times with high Compute Credit use and little progress
  • Large one-shot changes cause context loss or regressions
Recommended approach：
  1. Consult first： Use Consultation mode for plan and risk; switch to Execute now when clear. Consultation is free.
  2. Validate on real pages： Ask the AI to analyze your actual project:
  • “Do not change code yet. Using XX page in my project, analyze root cause and risks.”
  • “Repeat your understanding and fix steps; I will confirm before execute.”
  1. Small steps： One or two items per turn:
  • “Split this into 2 steps; only do step 1; I confirm before step 2.”
  • Fix login before list pages — avoid “fix every bug at once”
  1. Three-strike rule： After 3 failed attempts on the same issue:
  • Stop the current thread
  • Roll back to a working version
  • Consult with a new angle, or ask for a minimal repro before touching the main project:
  • “This failed 3 times. Explain the fix on a minimal example first; I confirm before changing the main project.”
  1. Troubleshoot first： Click Troubleshoot (free), then follow the debug flow below
Related： Credits guide → Money-saving checklist

​
Feature breaks again after a fix

Symptoms：
  • Fix works briefly then fails again
  • Same issue never resolves
  • High Compute Credit spend
Fixes：
  1. Rollback： Return to working version; try a different approach
  2. Consultation： Free consultation → plan → execute
  3. Avoid repeating the same prompt： Loops waste Compute Credits
  4. Complex features： For editors like poster tools, consider superun image generation skills
  5. Stop the thread： If the AI is in an error spiral, start fresh
Debug flow：
  1. Understand： Copy errors; ask what they mean and where the bug lives
  2. Confirm approach： Discuss fixes
  3. Then edit： Execute after alignment
💡 Tip： If the first attempt is wrong, the model may spiral — stop early and change approach.

​
Troubleshoot does not find the issue

Symptoms：
  • Bug or failed tests; Troubleshoot does not locate or fix it
Why：
  • Troubleshoot targets system, deploy, and environment issues (services, DB connection, redeploy)
  • Model misunderstanding or timing (consult OK, execute fails) often needs chat, not Troubleshoot
Fixes：
  1. In chat, state expected flow vs actual behavior
  2. Ask the AI to repeat understanding and steps before execute
  3. If still stuck, use the three-strike rule: rollback and rethink