Install a consistent set of working rules into a project so an AI coding assistant behaves the same way across sessions.
Pick a work-type profile such as software development, devops, marketing, or research to match your kind of project.
Run one task at a time through a plan, build, verify, ship cycle with the agent handling routine steps itself.
Set up the same configuration globally once, then add a thin per-project file for each new codebase.
| promptpartner/agentsmith | kelseyhightower/istio-ingress-tutorial | codecrafters-io/build-your-own-redis | |
|---|---|---|---|
| Stars | 313 | 321 | 327 |
| Language | Shell | Shell | Shell |
| Last pushed | — | 2018-07-26 | — |
| Maintenance | — | Dormant | — |
| Setup difficulty | moderate | hard | moderate |
| Complexity | 2/5 | 4/5 | 1/5 |
| Audience | developer | ops devops | developer |
Figures from each repo's GitHub metadata at analysis time.
Windows users need PowerShell 7+ to run the setup.ps1 equivalent.
Agentsmith is a setup system for AI coding assistants like Claude Code, Codex, and Gemini. The idea behind it is that quickly building a working demo is easy, but keeping a real project stable once customers show up, requirements change, or another engineer joins is a different problem, and that is the gap this project tries to close. It installs a set of rules, files, and habits into your project so an AI agent works more carefully and consistently over time, rather than acting differently every session. A single setup script, run with no options, launches a wizard that asks you questions such as which kind of work you are doing (software, devops, marketing, document writing, data work, research, design, or general admin), how cautious the assistant should be, and where the setup should apply: one project, your whole machine, or a portable copy you can move around. Based on your answers it builds a tailored configuration file and shows you the exact command before running anything, and any file it changes is backed up first. Power users can skip the wizard and pass the same options directly as command line flags. Once installed, the intended way of working is to take one tracked task at a time and move it through four stages: plan, build, check the work, then ship it, with the agent handling routine decisions itself and only stopping to ask a person about things like a missing password or an unexpected external problem. Nothing is meant to be treated as finished until real checks have run against it, and if something goes wrong the recommended habit is to fix the underlying rule so the same mistake is less likely to happen again. The project ships as a shell script, with a PowerShell version included for Windows users so the same options and behavior work on both platforms. There is also a bundled set of guide documents covering first-time use, safety settings, and troubleshooting. No license is stated in the portion of the README reviewed.
A configurable setup system that installs disciplined rules and workflow steps into a project so AI coding assistants work more consistently.
Mainly Shell. The stack also includes Shell, PowerShell.
Setup difficulty is rated moderate, with roughly 30min to a first successful run.
Mainly developer.
This repo across BitVibe Labs
Don't trust strangers blindly. Verify against the repo.