TangleGuard Claude Code Plugin

A Claude Code plugin that keeps your coding agent inside your architecture, powered by TangleGuard. It bundles:

• Skills that teach the agent to query the codebase's real architecture (layers, packages, dependency graph) instead of guessing it from file paths — and to preflight every import with check-import before writing it, so dependencies that violate your rules or close a cycle never enter the codebase. Before you commit, the changes skill shows you what the agent changed structurally — TangleGuard's own diff of the working tree against HEAD, verbatim, not the agent's summary of its work.
• A guardrail hook: when the agent finishes a task, it runs tangleguard validate --since HEAD and hands any rule violation or cycle the change introduced back to the agent, with file:line. Pre-existing issues are ignored, and a clean change produces no output. How it works.

Requirements

You need the TangleGuard CLI installed. The installation guide covers the install script, Homebrew and Windows.

Installation

This repo is a Claude Code plugin and its own marketplace:

/plugin marketplace add TangleGuard/claude-code-plugin
/plugin install tangleguard@tangleguard

Privacy

The skills and the hook run the TangleGuard CLI locally, and your code never leaves your machine. The CLI sends one anonymous usage event per command to OpenPanel: the command name, the CLI version, the operating system, and whether validate --since was used. We need it to know how the CLI is used. No code, file paths or user identifiers are sent. The CLI command tangleguard upload sends architecture data to TangleGuard Cloud (if the user has an account and provided an API token), but the plugin never runs the upload command. Details are in our privacy policy.

Other agents

For any agent (and for humans), the guardrail check works as a git pre-commit hook. Save this as .git/hooks/pre-commit and make it executable:

#!/bin/sh
exec tangleguard -q validate --since HEAD

The skills follow the Agent Skills format — point your agent's skill loader at the skills/ directory (architecture, validate, changes, untangle, assign-layers), or copy it into your project's skills folder (e.g. .claude/skills/, .codex/skills/).