autoresearch

Automated iterative code-quality improvement is achieved by measuring metrics, scaffolding experiments, and executing autonomous iterations that only keep changes if the measured metrics improve. You can scaffold a loop with the --scaffold command and run it using --run loop-name.

Autonomous code-quality experimentation is an agent-driven loop that performs controlled code changes to reduce specific quality metrics like complexity or lint failures. It uses Git rollback to automatically revert any changes that do not deterministically improve the defined metrics.

To scaffold and run an autonomous loop, use the --scaffold command to generate the necessary file structures, write your constraints in program.md, and execute the loop with --run loop-name. The loop repeats until your target quality metric is reached.

Yes, running autonomous code-quality iterations requires a defined skill contract including a frontmatter name and description in SKILL.md, a measure script under scripts/autoresearch, and a program.md file encoding the constraints for the evaluation loop.

Automated experimentation can target various code-quality metrics for steady reduction, including code complexity, duplication, churn, and lint violations. The agent-driven loop measures these metrics deterministically and keeps changes only when the chosen metric improves.

You should avoid using an autonomous agent-driven loop when your project lacks a deterministic measure script for quality metrics or cannot support Git rollback for reverting failed experiments, as the loop relies entirely on these mechanisms to evaluate and retain changes.