Fix production errors

Point at the error, let it fix

After /sentry writes a report, you have a list of issues ranked by severity. Some are one-line fixes. Some need a database migration. You could go through them one at a time, or you could let agents handle it.

Inside Claude Code, type /fix-issue. It reads the latest Sentry report, shows you what's pending, and asks which issues to fix. Pick specific ones by number, say "all critical," or just say "all."

How it works

For each issue you select, the system figures out which agent should handle it. A database error goes to the database agent. An API crash goes to the backend agent. A rendering bug goes to the designer. Auth issues get both backend and connector agents since auth spans layers.

Each agent gets the stack trace, the root cause analysis from the Sentry report, and the relevant feature documentation. It reads the actual file, understands the context, and applies a surgical fix. No refactoring the surrounding code, no "while I'm here" improvements.

If multiple issues can be fixed in parallel (different files, different features), they run simultaneously. If issues overlap on the same files, they run sequentially so fixes don't conflict.

After the fixes

Type checks run automatically. If the fix introduces a type error, a fixer agent cleans it up. The Sentry report gets updated with what was fixed, which agent did it, and which files changed.

/fix-issue                    # interactive: pick from the list
/fix-issue --all-critical     # auto-fix critical and high
/fix-issue --issue PROJ-123   # fix one specific issue

For unattended fixing with full guardrails (branch, tests, confidence scoring), use /heal instead.