autoresearch

Automate iterative keep/discard experiment loops over a configured repository.

11|1|Updated Feb 23, 2026

Search Tags:#automation #git #workflow #experimentation #experiment #autoresearch

Ready to run

npx skills add https://github.com/jimezsa/opencolab --skill autoresearch-jimezsa

💬 Prompt for ChatGPT / Claude Chat▼

Please help me install this Agent Skill.
Skill: autoresearch
Source: https://github.com/jimezsa/opencolab/tree/main/projects/SKILLS/autoresearch
Command: npx skills add https://github.com/jimezsa/opencolab --skill autoresearch-jimezsa

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Iterative keep/discard experiment loops over a single, explicitly configured repository, enabling disciplined, human-in-the-loop experimentation without manual drift.

Core Features & Use Cases

Bounded, repeatable experiment cycles with clear guardrails and branch management.
Enforces edits only to the configured file and confines operations to the target repo.
Provides structured reporting of outcomes to inform keeps/discards and next steps.

Quick Start

Configure the repo_path, editable_file_path, run_command, and metric_rule, then run a single bounded autoresearch loop to evaluate results.

Frequently Asked Questions about autoresearch

High-intent search queries and answers about installing and using this skill.

FAQPage Schema

How do I automate iterative experiment loops in a git repository?▼

You can automate iterative experiment loops by configuring a repository path, a single editable file, a run command, and a metric rule to execute bounded cycles and record outcomes.

What is the best way to manage keep or discard decisions for code experiments?▼

Managing keep or discard decisions is handled by running one experiment at a time over disposable branches, enforcing repo confinement, and recording structured outcomes to inform next steps.

Can I run multiple experiments simultaneously across different files in my repo?▼

No, you cannot run multiple experiments simultaneously; the automation enforces running one experiment at a time and restricts edits to a single configured editable file.

Do I need to provide a results file to evaluate experiment outcomes?▼

Providing a results file is necessary for recording outcomes; the automation evaluates the configured metric rule and writes the experiment results to the specified file when provided.

What are the limitations of using automated experimentation workflows?▼

Limitations include strict repo confinement, restriction to a single editable file, running only one experiment at a time, and requiring disposable branches for controlled bounded cycles.