Codex Autoresearch

Improve code through repeatable experiments.

Install - Try it - How it works - Dashboard - Docs

Codex Autoresearch helps improve local code against a repeatable benchmark. Give it a workload, correctness checks, an edit boundary, and a time budget. It records the baseline, evaluates small changes, and leaves a reviewable patch with the evidence behind it.

Use it for bounded performance or resource-use experiments: faster tests, lower build time, less memory, or higher throughput on a defined workload. Reviews, documentation, product research, and one-off fixes normally stay ordinary Codex work.

Version 3.0 also supports investigations where you know the outcome you want but have not settled on a method. You accept the objective, allowed changes, evidence requirements, and budget once. Preparation, failed attempts, repairs, confirmation, and delivery all use that same allowance. Goals without a numeric metric can use explicit conditions and counterexamples. See Bounded investigations, or use the short benchmark path below.

The 3.0 release is based on engineering verification. No model-driven comparison has been run, so it makes no claim of better outcomes or lower cost than ordinary Codex or 2.9.0. The optional comparison protocol and collection harness stays disabled unless a study receives a separate budget.

The screenshot uses synthetic example data.

The loop is inspired by karpathy/autoresearch and pi-autoresearch. This version is built around Codex, local repositories, and ordinary reviewable Git work.

Install

Open the plugin picker in Codex:

/plugins

Choose TheGreenCedar -> codex-autoresearch -> Install plugin, then start a new Codex task in the repository you want to improve.

If your Codex build supports marketplace management from the terminal, you can register the source marketplace first:

codex plugin marketplace add TheGreenCedar/AgentPluginMarketplace --ref main

The marketplace lives in TheGreenCedar/AgentPluginMarketplace; this repository is the plugin source.

Try it

After installing the plugin, open Codex in the repository you want to improve and give it a goal, a benchmark, and a boundary. Use this as a starting point for the contract:

/goal @Codex Autoresearch run a measured loop to reduce parser runtime.
Benchmark: node bench/parser.mjs
Metric: seconds (s), lower is better
Checks: node --test tests/parser.test.mjs
Scope: src/parser.mjs
Protect bench/parser.mjs and tests/parser.test.mjs.
Measure two baselines and repeat each candidate before a keep.
Stop after 5 packets or 30 minutes; repeat measurements count toward that limit.
Propose the complete contract for approval before setup or execution.

Codex first presents the workload, checks, scope, and budget for review. After you accept the contract, Autoresearch measures the baseline before any candidate change. Commands and paths in the example must match your project.

If you do not know what the benchmark should be, say what outcome you want and ask Codex to propose one: