autoresearch

Automated iterative code experimentation modifies code, runs a user-defined metric command, and keeps only changes that measurably improve the numeric metric while reverting regressions. You provide the exact command, extraction rule, and target files to optimize metrics like runtime or bundle size.

Autonomous code optimization creates a git branch, commits experimental changes, runs your measurement command, and extracts numeric results to keep improvements and revert regressions. It logs each attempt to results.tsv, automating performance tuning by systematically testing code modifications against a baseline metric.

Yes, autonomous experimentation works on any git-backed codebase where you can provide a reproducible command outputting a numeric metric like binary size. It requires terminal and git access, permission to modify in-scope files, and the ability to create commits and branches while logging results.

Interactive setup walks you through defining optimization goals, exact metric commands, extraction rules, in-scope and out-of-scope files, constraints, and experiment budgets. Safe guardrails respect out-of-scope files, require approval for environment changes, enforce baseline measurements, and limit experiments to your constraints.

Autonomous iterative experimentation requires a reproducible command producing a numeric metric, limiting use cases where measurements cannot be automated. It also requires explicit user approval for environment changes and cannot modify out-of-scope files, constraining optimizations to defined boundaries.

Yes, autonomous code optimization requires terminal and git access to create branches and commits. You must also provide an explicit metric command with an extraction rule, have permission to modify in-scope files, and be able to log experiment results throughout the optimization loop.