Topic
The bug came from a malformed record nobody typed by hand. The agent fixed the parser, wrote a clean sample that parsed, watched it pass, and marked the edge case handled. It never replayed the record that started the thread. A check you authored yourself grades itself.
Read articleThe agent reads a file once and keeps that buffer across a long session. When the disk changes, the memory does not: it quotes old text, argues against a version nobody shipped, and patches stale content. The gate: invalidate the cache and re-read before editing.
Read articleThe agent minimized a failing request to isolate the bug, and somewhere in the trimming it removed the exact field that triggered the branch. The tiny repro passed, so it reported fixed; the original payload still crashed.
Read articleCI stayed green after the agent added a token-bucket test, so it reported done. Run that file alone and it fails: an earlier test left state in a shared registry, and the new case read leftover data instead of building its own. Fixed sequential order hides this order pollution.
Read articleThe suite stayed green after a package rename, CI was green, and the changelog promised formatBudget as a public export. The first consumer to upgrade hit formatBudget is not a function at runtime: tests ran against TypeScript source through a path alias and never touched the built dist barrel.
Read articleThe import finished, logged 'processed successfully,' and returned zero errors. About a third of the source rows had never landed — the parser skipped records whose shape differed from the happy example and said nothing.
Read articleThe agent edited a handler, re-ran the request, saw the old response, and still wrote done in the summary. The long-lived dev process had never reloaded the shared module, so it verified against code that was already replaced on disk.
Read articleThe agent renamed the class, updated every reference, and CI went green. The thrown error still said the old name, and two weeks later a support ticket traced the confusion to a string the symbol-graph rename never touched.
Read articleThe agent guarded the user-facing call site and the test went green, but the same helper sat behind an internal batch job the fix never touched. Two weeks later the sparse export blew up on the exact null the patch was supposed to kill.
Read articleAn hour earlier we had chosen the retry queue over a direct write. Switching editors to run the migration, the fresh session saw no recorded decision and calmly proposed the direct write again. Forty minutes later, the closed decision had been reopened and compromised.
Read articleThe ticket was a failing validation on an invalid webhook payload. CI went green, but the diff did not fix the field — it stopped raising the error. The invalid case returned 200 and vanished into a log line, because no test asserted the expected failure.
Read articleTwo tickets, two worktrees, both green. The merge kept the second agent's hunk in the shared registry and dropped the first agent's flag; no test failed, because neither suite asserted the shared file's full contents.
Read article