Find and fix a bug

Review with evidence and counterexamples

A finding needs a defect, location and reproduction. A read-only review cannot claim test execution. Verify the new revision after a fix and keep unknowns explicit.

A repair report needs a reproduced fault and a check after repair.

Review the repaired version

A finding needs a location, reproduction and impact. “Long function” alone does not prove a defect. Rerun the check on the repaired version.

A repair report should name the fault and the check after repair.

Practice

Fix case- and accent-insensitive search.

Reproduce may in failing to find T01. Edit only findTickets in model.mjs for case/accent-insensitive substring search, including đ/d; trim query and preserve order/seed. Run search, then build/access tests.

Check your work

Before/after reproduction evidence is recorded.

Search and earlier checks still pass.

References

• Verifying unsupervised runs — Claude Academy
• Claude subagents — Claude Code Docs
• OpenAI iterative repair loop — OpenAI Docs
• Anthropic agent evaluations — Anthropic Engineering

• Back to the AI Builder roadmap
• Understand AI agents
• Shape the idea
• Set up your workspace
• Give a clear task
• Keep project rules clear
• Design a usable journey
• Build the first feature
• Check access rules
• Test beyond the happy path
• Prepare a release
• Observe and respond
• Add tools when needed
• Split work and hand it over