self-improve

Autonomous codebase performance optimization uses an evolutionary loop applying research, planning, execution, and tournament selection via benchmarking. It runs fully autonomously after an initial gate without pausing for user confirmation, continuously improving measurable performance goals safely using git worktrees.

Tournament selection for evolutionary code improvement benchmarks competing code variations against measurable goals like performance or accuracy. The system selects winning variations to proceed while enforcing harness rules that prevent benchmark tampering, ensuring safe and measurable evolutionary optimization.

Applying git worktree-based changes for autonomous refactoring isolates experimental code variations in separate worktrees to protect the main branch. The system enforces safe constraint rules via a benchmarking harness and maintains persistent state under a dedicated directory for crash resumability.

Resuming an interrupted code quality improvement loop is fully supported via persistent state stored under a dedicated directory. This resumability allows the evolutionary optimization process to recover and continue executing benchmarking safely after crashes or interruptions without losing prior progress.

Limitations of autonomous evolutionary optimization include its restriction to measurable code-quality goals like performance and accuracy improvements, requiring a benchmark harness. The no-interruption execution policy prevents pausing for user confirmation, and strict harness rules are required to enforce constraints and prevent benchmark tampering.

Benchmark harnessing for iterative code refactoring is needed when applying evolutionary optimization to achieve measurable performance or accuracy goals. It provides the tournament selection framework to evaluate code changes objectively and enforces safe constraint rules preventing benchmark tampering during autonomous execution.