sherlock

You can fix flaky tests and memory leaks by running an autonomous investigation loop that establishes baselines, runs experiments, measures results, and keeps or reverts changes based on evidence. This loop requires a Git repository with standard project metadata.

Automating reliability improvements in CI workflows involves running an autonomous health-improvement loop that iterates on performance issues and intermittent errors. It creates branches, runs baseline tests, and logs results to a dedicated .claude/sherlock directory for traceability.

Yes, you can scope the autonomous investigation to focus on specific files, directories, tests, performance, or memory issues. This targeted approach ensures the health-improvement loop only runs experiments and measures results in the areas you specify.

Autonomous investigation requires a Git repository with standard project metadata to function correctly. It uses this setup to create branches, establish baselines, run test experiments, and log outcomes to a dedicated directory for ongoing traceability.

The autonomous investigation loop decides to keep or revert code changes by measuring experiment results against established baselines. If a change resolves a performance issue or reliability problem without regressions, it is kept; otherwise, it is reverted.

Performance fixes may not persist if they are not validated through a baseline-driven remediation loop. An autonomous investigation establishes baselines, measures results against them, and logs outcomes to a dedicated directory to ensure only evidence-backed changes are kept.