fix-failing-test

Fix a deliberately introduced failing test. Category: testing. Scope: small.

omot bench start fix-failing-test --param file=src/lib/math.ts --agent <your-agent>

ParametersCopy Anchor Link

PreconditionsCopy Anchor Link

• A clean working tree and a test script (scripts.test) that passes at the baseline.
• The file is a non-test source file whose tests catch a flipped operator.
• You confirm that omot may create a throwaway branch and commit the bug there.

OutcomesCopy Anchor Link

• pass: The test script passes, no test file was deleted or modified, and no tests were skipped.
• partial: The test script passes, but test files were modified.
• fail: The test script fails, a test file was deleted, or tests were skipped.

How it is verifiedCopy Anchor Link

• bench start requires a clean working tree and a passing test script; a baseline failure blocks the task and reports the output tail. After you confirm (or pass --yes), it creates the branch omot/fix-failing-test-<timestamp>, flips one operator in the file (for example >= to >), and records the changed tree in a single parentless commit named bench baseline. The commit has no parent, so its history exposes no diff; the original branch commit and real index are untouched. This is not a security boundary: an agent can still inspect other refs such as origin/main or the reflog.
• The prompt does not name the file, so the agent has to find the bug.
• bench verify runs the test script and compares test files with the baseline: deleted test files or new .skip/.only/.todo/xit markers fail the run, and any other test file edit makes it partial.
• Every verification is appended to .omot/run.json, with its own commit and result; the latest result records how many attempts were made.
• When you are done, switch back with git switch - and delete the branch with git branch -D omot/fix-failing-test-<timestamp>.

The prompt and verifier live in packages/benchmarks/src/tasks/fix-failing-test. Changing either requires a version bump in task.json.