ClaudeForge plugin
A Claude Code plugin that generates and maintains CLAUDE.md: a 0 to 100 quality score, a 150-line cap and stale reference checks
- 430 ★ on the repository
- Last commit
Extended access
Why this level
- Hooks run Python scripts on every file write and edit and whenever instructions load
- The commands create, split and overwrite CLAUDE.md and AGENTS.md files in the project, and the script install executes code downloaded from the network
Install
- Works in
Manual install
/plugin marketplace add alirezarezvani/ClaudeForgeFirst step: add the plugin marketplace in a Claude Code session.
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- Hooks run Python scripts on every file write and edit and whenever instructions load
- The commands create, split and overwrite CLAUDE.md and AGENTS.md files in the project, and the script install executes code downloaded from the network
This is third-party code. Review the repository files before installing.
What it does
ClaudeForge creates, improves and maintains CLAUDE.md files in Claude Code projects. The /enhance-claude-md command explores the repository through the Explore subagent, detects the stack, project type and team size, shows what it found and writes files after your confirmation. A multi-part project gets a root CLAUDE.md plus nested ones for backend and frontend. For an existing file the skill gives a score from 0 to 100: length, section completeness, formatting, specificity and modularity. A hook on the PostToolUse and InstructionsLoaded events holds the 150-line cap per file: on a violation the validation script returns an error, and a large project is spread across sub-files through @path imports. The /sync-claude-md command walks every CLAUDE.md, prunes stale references, splits files that are too long and repairs the chains. With the --weekly flag it runs three audits in parallel: claude-md-drift-audit, claude-md-link-check and claude-md-dependency-rescan. The /claude-to-agents command turns the CLAUDE.md tree into an AGENTS.md for Codex, Gemini Code Assist and other tools: as a symlink, a copy or a single file with the import chain inlined. The claude-md-guardian agent runs on the haiku model, keeps the file in step with the code, never commits on its own and aborts when there is no validated output.
Who it is for. For developers and teams who run projects in Claude Code and want CLAUDE.md files that stay short, current and in line with Anthropic best practices for Claude Code.
Good fit when
- The project has no CLAUDE.md: /enhance-claude-md explores the repository and creates a file for your stack
- CLAUDE.md has grown, gone stale or points at deleted paths: /sync-claude-md prunes the excess and splits the file
- AGENTS.md, .cursorrules or .windsurfrules sit alongside: the command chains them through imports instead of overwriting
- You need an AGENTS.md for Codex or Gemini from the CLAUDE.md tree: /claude-to-agents does it
- Once a week you want an audit of drift from the code, broken links and dependencies: /sync-claude-md --weekly
Not a fit when
- You need rules for one task and no permanent file: the plugin is built for long-lived CLAUDE.md files
- You do not work in Claude Code: the plugin hooks and commands are tied to it
- The hard cap gets in the way: the hook rejects files over 150 lines, and only *.local.md is exempt
- You are not ready to run an install script that is downloaded and executed at once: there is an install through /plugin
Example request
Run /enhance-claude-md and create a CLAUDE.md for my TypeScript, React and PostgreSQL project; show the detected stack and file structure firstLimitations
Claude Code 2.1.4 or newer is required. Hooks check the 150-line cap per CLAUDE.md, so large projects have to be split into sub-files with @path imports. The CLAUDE.local.md file for personal settings has no cap and is gitignored automatically. Codex and Gemini do not resolve @ imports, so for modular projects the --inline-chain mode is recommended when exporting to AGENTS.md. Four Karpathy principles are embedded in every generated CLAUDE.md: think before coding, simplicity first, surgical changes and goal-driven execution. The main skill and the guardian agent run on the haiku model.
How to disable. Remove the plugin through the /plugin menu or, if you used the script, the files install.sh put in place. On your own machine the hooks can be turned off in hooks/hooks-config.local.json, which stays out of git.
FAQ
How do I install ClaudeForge for Claude Code?
The plugin installs with two /plugin commands inside a Claude Code session: adding the marketplace and installing. The plugin installs skills, commands and the guardian agent, and at the user level they are available in every project. After a restart call /enhance-claude-md.
How does the 150-line cap work and how do I get around it?
A hook checks the file at write time and at instruction load time and returns an error if it has more than 150 lines. A large project is split into sub-files with @path imports. *.local.md files fall outside the cap. You can turn the hook off just for yourself in hooks-config.local.json.
How do I use ClaudeForge together with Codex and Gemini?
The /claude-to-agents command builds an AGENTS.md from the CLAUDE.md tree in one of three modes: --symlink (default on macOS and Linux), --copy and --inline-chain. For modular projects --inline-chain is recommended because Codex and Gemini do not resolve @ imports. An existing AGENTS.md is backed up before overwriting.
What do the three audit skills check?
Drift-audit walks recent git history and flags lines that point at deleted or renamed paths and removed dependencies. Link-check verifies that every @path import and relative link in each CLAUDE.md resolves. Dependency-rescan diffs the dependencies from package.json, requirements.txt, pyproject.toml, go.mod and Cargo.toml against the Tech Stack section. All three are read-only.
Can the claude-md-guardian agent commit anything?
No. It aborts when there is no validated output and never commits on its own. It respects the local hooks-config.local.json file as well.
Discussion
No questions or reviews yet.
Related
PluginsA skills library that gives coding agents a development process: brainstorming, planning, TDD, subagents and code review
Skills for real engineers by Matt Pocock
Skills For Real Engineers
Small composable skills for engineering with agents: plan grilling, TDD, bug diagnosis, code review and architecture
Karpathy-inspired Claude Code guidelines
Karpathy-Inspired Claude Code Guidelines
Four principles in one CLAUDE.md against common LLM coding mistakes: silent assumptions, overengineering, drive-by edits
A plugin and skill that makes the agent write minimal code: reuse, stdlib and native platform features come first
GitHub toolkit for spec-driven development: the specify CLI adds agent commands and skills to a project, from principles to implementation
Garry Tan's skill suite that turns Claude Code into a virtual team: product, architecture, design, review, QA and shipping
Comparison with similar tools
Entries that share the main search query, and how they differ.
| Entry | Official | Repository stars | Main difference |
|---|---|---|---|
| ClaudeForge (this entry) | no | 430 | |
| Caliber | no | 1,310 | a command-line tool that generates and maintains context files for several agents at once: CLAUDE.md, Cursor rules, AGENTS.md and Copilot instructions |
| agnix | yes | 457 | a linter for CLAUDE.md, AGENTS.md, SKILL.md, hooks and MCP configs with autofix |
README in short
The page opens with a block of version two news: the plugin manifest, the hard cap, the sync command, Karpathy principles and layered hook configuration. It then lists the contents (five skills, three commands, an agent, hooks), install options, scenarios for a new and an existing project, documentation and examples.