Archcore is a spec-driven development and Git-native context engineering tool for AI coding agents. The core problem it solves is this: although coding agents can write code, they do not understand your project—what a feature should do, where the code should go, and which decisions and rules are already in effect. So the agent guesses, and you have to explain the same things again in the next session. Archcore keeps specs, architecture, decisions, rules, and plans in Git and provides the right project context while the AI coding agent works, helping the agent make changes that align with the repository's architecture, rules, and past decisions.
Main capabilities Archcore provides three core commands that can be used individually or combined for the same change:
-
/archcore:plan: Plan before building. The agent first reads the repository and existing documentation, then asks only for information it cannot find in the documentation. Your answers become documents in .archcore/: a spec containing requirements, and a plan that maps tasks to files. User-facing changes also get Given/When/Then examples. Larger work can add a PRD, research, or a formal requirements chain. Archcore weighs the change, computes the route, and reports a size from S to XL; you do not need to choose a template or size. Risk raises the size: security requirements add a formal requirements chain, and data migrations add a migration runbook. The agent then implements based on the spec, examples, and tasks, with open questions resolved before the code rather than after.
-
/archcore:document: Document as you speak. One sentence from you becomes a complete, linked document. /archcore:document records what is currently true: a decision, a team standard, how a module works, or an operational guide. You do not need to choose a format; Archcore selects the document type, reads the code and existing documentation, and checks whether the document already exists. It asks only for information it cannot find and links the new document to related documents. A decision can also produce accompanying rules and guidelines. If there is no topic, /archcore:document reads the changes on the branch and asks one question about what to document. Documentation outlives sessions; it is stored in Git alongside the code, has status (draft first, then accepted), and is checked against the code during review.
-
/archcore:review: Review before merging. When documentation covers the change, /archcore:The review compares the branch and the documentation in both directions. Each finding points out the inconsistent code and documentation and gives a verdict: code-wrong (the code breaks documentation that is still valid) or spec-wrong (the documentation is out of date). You fix the correct side. When a plan covers the branch, the review checks its tasks and closes the plan when the work is done. When no documentation covers the change, the review still checks the branch against the project's existing decisions and rules. If the branch repeats an undocumented pattern, the review proposes documenting it. To record work shipped without a plan, run /archcore:document.
Usage On macOS, Linux, or WSL, and on Windows (PowerShell 5.1+), the installer adds the CLI and plugins for Claude Code, Codex CLI, and GitHub Copilot CLI (if found). archcore init connects the agent of your choice to the current project. It also supports Gemini CLI, OpenCode, Roo Code, and Cline. Cursor requires an extra setup step. If you already have CLAUDE.md, AGENTS.md, or rule files, you can type /archcore:init import in the agent to convert them into project documentation. The documentation stays in your repository, in the .archcore/ directory.
Slash commands work in Claude Code, Cursor, Codex CLI, and GitHub Copilot. In every connected agent, plain sentences also work, for example "Plan [your feature]", "Record why we chose [X]", "Review my branch".
When to use it Archcore is for teams or individual developers who want AI coding agents to follow the project's architecture, rules, and past decisions. It is especially suited to spec-driven development, context engineering, and scenarios where you want project knowledge versioned as documentation alongside the code. By turning project knowledge into files, every command leaves plain Markdown in .archcore/, versioned alongside the code it describes. The document type is reflected in the file name. Specs are only part of the context; decisions, rules, plans, and guides coexist with them. Changes to documentation are diffs in pull requestsjust like changes to code.
Limitations Archcore requires use with a supported AI coding agent, including Claude Code, Cursor, Codex CLI, GitHub Copilot, Gemini CLI, OpenCode, Roo Code, and Cline. Cursor requires an extra setup step. Documentation is stored in the .archcore/ directory and requires a Git repository. For details about privacy, refer to the project documentation. Archcore is licensed under Apache 2.0, and the source code is on the dev branch.
