No forms. Pick the sentence that sounds like your week.
Teach every assistant your conventions, your gotchas, your review bar — once — and stop re-explaining them in every prompt.
For engineers →Define behavior at the org level, deliver it to every repo by pull request, and see when a repo drifts from what you shipped.
For your org →Point one repo at the hub. Nothing touches your personal setup. You can read exactly what changes before it does.
Connect a repo →AgentBoot is a build tool, not a runtime. It produces files and exits — nothing runs alongside your agents.
Write traits, personas, and rules as markdown in one hub repo. Reusable behavioral building blocks, composed with a single config file.
`agentboot build` resolves references and emits native config for each supported tool — Claude Code, OpenAI Codex, GitHub Copilot — from one source.
`agentboot sync` writes those files to every target repo as a pull request. Drift detection flags any repo that no longer matches what you shipped.
You can — until you have more than one repo and more than one tool. Here’s where each shortcut stops working.
AgentBoot is what your team runs — your personal AI setup is untouched. Ready-made personas for code review, security analysis, tests, and data. Ask /ab in natural language; it scans before it changes anything.
Define behavior at the org level and deliver it to every repo by pull request. Scope hierarchy — org → group → team → repo — lets teams customize without escaping policy. Drift detection shows you when a repo no longer matches what you shipped.
The same three questions a security reviewer already asks — answered by pull requests and hashes, not promises.
Author behavior and policy once, in one hub repo — versioned and reviewable like any other code.
Emit blocking pre-tool-use hooks and managed settings to the CLIs that support them — the rule fires at the tool boundary rather than asking the agent to comply. Depth varies by platform; the matrix below says where.
Drift detection compares each repo against what you shipped and flags what no longer matches. Every delivery is a reviewable PR.
| Capability | Claude Code | Codex CLI | Copilot CLI |
|---|---|---|---|
| Compiled instructions | ✓ | ✓ | ✓ |
| Blocking pre-tool-use hooks | ✓ | ✓ | ✓ * |
| Managed settings | ✓ | — | — |
| Drift detection | ✓ | ✓ | ✓ |
Official support today covers each tool’s CLI surface. Broader surface support — IDE and editor extensions, and additional platforms — is on the roadmap. * Copilot’s hook surface is narrower than Claude Code’s and Codex’s — real, but lower-ceiling, and its exit-2 blocking is documented rather than yet empirically verified; we say so on purpose. Cursor, Windsurf, Gemini, and JetBrains are supported at a community tier — advisory guidance, not an enforced control. If a cell says advisory, treat it as advisory.
The honest boundary — because the last thing you should read before installing is where the tool stops.
No AI-provider account needed for core features.
Self-hosted, Apache-2.0, plain files out. Your prompts never leave your machine.