Repo Agent Kit is a repository instruction file tool for coding agents, focused on being free and privacy-first, running entirely locally in the browser.
Purpose It solves the core problem of which AI agent instruction files a repository actually needs. Different coding tools (Codex, Claude Code, Cursor, Copilot, Gemini CLI, Devin Desktop) each use different instruction file names and structures, and users often need to repeatedly copy the same rules into every tool. Repo Agent Kit helps users select the tools they use and provides the minimal and correct set of files, using AGENTS.md as the shared source to avoid duplicate maintenance.
Main capabilities
- Tool selection and file plan: After selecting Codex, Claude Code, Cursor, Copilot, Gemini CLI, or Devin Desktop, the recommended plan updates immediately, providing the minimal required file set (for example, 6 tools require only 4 files).
- Generate an AGENTS.md draft: You can create a draft of a standard instruction file locally in the browser, then connect it to other coding tools based on the plan.
- Generate from a manifest: If you already have package.json or pyproject.toml, you can drag in a manifest file to locally generate a directly copyable AGENTS.md containing detected commands, with no CLI, upload, or API key required.
- Repository audit: You can audit the instruction file readiness of public GitHub repositories, generate a shareable report, and create evidence-based instruction files without an AI key.
- Instruction content coverage: Helps turn vague preferences into agent-executable instructions, including preserving architecture (telling what can be reused, what should not be rewritten, and the scope of changes), incorporating testing, checks, accessibility, and final handoff into tasks, and protecting the repository (avoiding accidental harm to user changes, credentials, lock files, and migrations).
How to use Users first select the coding tools they use, and the system immediately provides the recommended file plan. If the standard instruction file itself is needed, a draft can be generated locally in the browser. If package.json or pyproject.toml is already available, the manifest file can be dragged in to generate AGENTS.md locally. The audit feature can inspect public repositories and generate a shareable report. Throughout the process, project details are not uploaded, and the generator builds and downloads files entirely in the browser.
Applicable scenarios Suitable for development teams or individuals using multiple coding agentswho want to manage repository instructions uniformly with a minimal file set; it is also suitable for scenarios where AGENTS.md needs to be quickly generated for an existing repository, or where the instruction file readiness of a public repository needs to be assessed.
Limitations The current generator runs entirely in the browser and does not upload data. The audit feature only checks public files, does not require GitHub login, and never requests private repositories. More information about fees, pricing, platform support scope, and similar details is not yet provided in the materials.
