Traction scoreGitHub stars can be faked, so popularity alone can be misleading. Traction Score looks for broader signs of recent attention, adoption, and active maintenance.
Autoresearch provides a structured framework for goal-directed iteration. It enables AI agents to autonomously modify code and verify the results, keeping or discarding changes based on defined performance metrics.
This approach helps automation builders implement modify-verify loops, turning repetitive optimization tasks into a systematic process of autonomous refinement.
Compatible AgentsThe repository documents support for these agents. The skills may also work with other agents that can load SKILL.md files, but they may need some setup or small changes.
Installs: A project-level copy of the Autoresearch skill and its command bundle. · Claude Code
Before you start
Claude Code installed.
Git available in the terminal.
Platform: macOS, Linux, or native Windows with Git Bash.
Run from: The target project root.
From the target project root, run this complete block in a terminal. It clones the repository into `autoresearch` and copies the skill and commands into the project's `.claude` directories.
Installs: A global copy of the Autoresearch skill and its command bundle. · Claude Code
Before you start
Claude Code installed.
Git available in the terminal.
Platform: macOS, Linux, or native Windows with Git Bash.
Run this complete block in a terminal from a directory where the repository may be cloned as `autoresearch`. It copies the skill and commands into the global Claude Code directories.
Installs: The global OpenCode Autoresearch package and command bundle. · OpenCode
Before you start
OpenCode installed.
Git and Bash available in the terminal.
Platform: macOS, Linux, or native Windows with Git Bash.
Run this complete block in a terminal. It clones the repository, enters the checkout, and invokes the guided installer for a global OpenCode installation.
git clone https://github.com/uditgoenka/autoresearch.git
cd autoresearch
./scripts/install.sh --opencode --global
Installs: A project-level copy of the Autoresearch skill and its OpenCode command bundle. · OpenCode
Before you start
OpenCode installed.
Git available in the terminal.
Platform: macOS, Linux, or native Windows with Git Bash.
Run from: The target project root.
From the target project root, run this complete block in a terminal. It clones the repository and copies the skill and command files into the project's `.opencode` directories.
Installs: The global Codex Autoresearch package, including its bundled orchestrator and regression helpers. · OpenAI Codex
Before you start
OpenAI Codex installed.
Git and Bash available in the terminal.
Platform: macOS, Linux, or native Windows with Git Bash.
Run this complete block in a terminal. It clones the repository, enters the checkout, and invokes the guided installer for a global Codex installation.
git clone https://github.com/uditgoenka/autoresearch.git
cd autoresearch
./scripts/install.sh --codex --global
Installs: A global copy of the Codex Autoresearch skill. · OpenAI Codex
Before you start
OpenAI Codex installed.
Git available in the terminal.
Platform: macOS, Linux, or native Windows with Git Bash.
Run this complete block in a terminal from a directory where the repository may be cloned as `autoresearch`. It copies the Codex skill into the global Codex skills directory.
Treat `72%`, `90%`, `src/**/*.test.ts`, `src/**/*.ts`, and the shown verification command as samples; replace them with the project's actual baseline, target, allowed file globs, and mechanical coverage command.
/autoresearch
Goal: Increase test coverage from 72% to 90%
Scope: src/**/*.test.ts, src/**/*.ts
Metric: coverage % (higher is better)
Verify: npm test -- --coverage | grep "All files"
Iterations: 25
Turn a goal into a validated experiment configuration
Uses autoresearch
A concrete improvement goal to replace `[goal]`.
The project directory or relevant files to replace `[project directory or relevant files]`.
Any known test, benchmark, lint, build, or measurement command that could provide mechanical verification.
Use the autoresearch skill's `/autoresearch_plan` mode to convert this goal into a validated Scope, Metric, and Verify configuration: [goal]. Use [project directory or relevant files] as the available context. Return the proposed configuration for review; do not start the iteration loop.
Investigate a reproducible bug with bounded iterations
Uses autoresearch
The failing test, error text, or reproducible behavior.
The files or directories that may be examined or changed.
An existing reproduction command or exact reproduction steps.
Use the autoresearch skill with `$autoresearch debug` to investigate [failing test, error, or reproducible behavior] within [allowed files or directories]. Use [reproduction command or exact reproduction steps] to test each hypothesis. Keep the investigation bounded to 5 iterations, report the hypotheses and findings, and do not push, publish, or deploy anything.
Claude
Codex
OpenCode