BrainHarness Autoresearch
Run controlled prompt experiments with explicit checks. Compare results before applying a change.
How to use it
- Prepare the target prompt and an eval.json with test inputs and pass/fail criteria.
- Choose a supported model provider, configure its API credentials and set an experiment budget.
- Run the experiment, inspect the results and review a candidate before applying it.
Example request
Which prompt actually works better?
Requirements and limits
Requires Python 3.10+ and model API access. Experiments may incur charges. Results depend on your evaluation criteria; prompt optimization does not fix code or architecture bugs.
Install
Claude Code
/plugin marketplace add zning1994/brainharness
/plugin install brainharness-autoresearch@brainharnessCodex
codex plugin marketplace add zning1994/brainharness
codex plugin add brainharness-autoresearch@brainharnessInstall from the repository marketplace. Autoresearch also requires Python and your model API credentials.
ClawHub
Skills are listed on ClawHub. Check the listing for the currently downloadable version. ClawHub
ChatGPT
Not yet listed in the public ChatGPT plugin directory. Codex compatibility does not imply public ChatGPT availability.
Common questions
Does it guarantee a better prompt?
No. It compares candidates against your tests. Use representative inputs, inspect failures and verify any change before adoption.