codex-autoresearch

Why we included this project

Most improvement requests in a codebase are really optimization problems: make the tests faster, shrink the bundle, cut memory use. This Codex plugin turns that vague goal into a loop of small experiments, each one measured against a baseline before anything changes. You give it a benchmark command that prints a metric, a set of correctness checks, and a scope boundary, and it records what happened with each attempt, keeping the useful results separate from the dead ends. The payoff is an evidence trail you can actually inspect: session history and reviewable Git branches, rather than a confident claim that things got faster.