AnchorMD

Generate optimized CLAUDE.md files for AI coding agents in seconds.

Stop hand-rolling CLAUDE.md. Let Forge analyze your codebase and generate a production-grade configuration file that makes Claude Code, Cursor, Windsurf, Codex, and OpenCode actually understand your project.

Why?

AI coding agents are only as good as the context you give them. A well-crafted CLAUDE.md is the difference between an agent that writes idiomatic code and one that fights your conventions on every change.

AnchorMD:

• Scans your codebase to detect languages, frameworks, and patterns
• Generates a complete CLAUDE.md with coding standards, commands, and anti-patterns
• Audits existing CLAUDE.md files and scores them against best practices
• Framework-aware presets for React, FastAPI, Rust, Django, Next.js, and more

Install

pip install anchormd

Quick Start

# Generate a CLAUDE.md for your project
anchormd generate .

# Audit an existing CLAUDE.md
anchormd audit ./CLAUDE.md

# Interactive setup
anchormd init .

# See what would change
anchormd diff .

# List available presets
anchormd presets

# List framework-specific presets
anchormd frameworks

Example Output

Running anchormd generate . on a FastAPI project produces:

# CLAUDE.md — my-api

## Project Overview
my-api — FastAPI service for account and billing workflows.

## Current State
- **Version**: 0.4.0
- **Language**: Python
- **Files**: 47 across 2 languages
- **Lines**: 3,204

## Tech Stack
- **Language**: Python
- **Framework**: fastapi
- **Package Manager**: pip
- **Linters**: ruff
- **Test Frameworks**: pytest
- **CI/CD**: GitHub Actions

## Coding Standards
- **Naming**: snake_case
- **Type Hints**: present
- **Docstrings**: google style
- **Imports**: absolute

## Common Commands
...

## Anti-Patterns (Do NOT Do)
- Do NOT use synchronous database calls in async endpoints
- Do NOT return raw dicts — use Pydantic response models
- Do NOT use `os.path` — use `pathlib.Path` everywhere
...

GitHub Action

Add automated CLAUDE.md auditing to your CI pipeline:

# .github/workflows/claudemd-audit.yml
name: Audit CLAUDE.md
on: [pull_request]

jobs:
  audit:
    runs-on: ubuntu-latest
    permissions:
      pull-requests: write
    steps:
      - uses: actions/checkout@v6
      - uses: Arete-Consortium/anchormd@v0.5.0
        with:
          fail-below: 40     # Minimum passing score (0-100)
          comment: true       # Post results as PR comment

The action posts a formatted comment on your PR with score, findings, and recommendations.

Free vs Pro vs Strict

Feature	Free	Pro $8/mo	Strict $49/seat/mo
generate — scan and produce CLAUDE.md	Yes	Yes	Yes
audit — score existing CLAUDE.md	Yes	Yes	Yes
verify, harvest, patch	Yes	Yes	Yes
11 community presets	Yes	Yes	Yes
init — interactive setup	-	Yes	Yes
diff — detect drift	-	Yes	Yes
6 premium presets	-	Yes	Yes
tech-debt, github-health, cleanup	-	Yes	Yes
Fail-closed validation (ANCHORMD_STRICT=1)	-	-	Yes
CI integration (PR comment automation)	-	-	Yes
Drift detection (generate, fix, LLM judge, HTML)	-	-	Yes
Fleet audit --json + history	-	-	Yes
Team seats (one key → N machines)	-	-	Yes
Audit log (1yr retention, CSV/JSON export)	-	-	Yes
99.5% license server SLA	-	-	Yes

Get Pro: Monthly ($8/mo) | Yearly ($69/yr)

Get Strict (for CI / teams / procurement-driven buyers): visit anchormd.dev/?page=strict — Seat Monthly $49/seat, Team-5 Annual $399, Team-25 Annual $1,490. Full feature documentation: docs/strict.md.

All 5 Tools Bundle: Monthly ($29/mo) | Yearly ($199/yr) — includes anchormd, agent-lint, ai-spend, promptctl, context-hygiene

Activate:

export ANCHORMD_LICENSE=ANMD-XXXX-XXXX-XXXX

Breaking in v0.6.0: ci_integration and advanced drift commands (drift generate, drift fix, drift report --ci, drift report --html, --judge-model) moved from Pro to Strict. Existing Pro subscribers on those features should subscribe to Strict or pin anchormd==0.5.0. See CHANGELOG for the migration guide.

Strict Mode (CI / Unattended)

anchormd fails open by default: if the license server is unreachable and no cache exists, a valid-format key grants Pro so a transient network blip does not break a working install. Strict subscribers can set ANCHORMD_STRICT=1 to reverse that — any validation failure (missing key, server unreachable with no cache, revoked or expired key) exits non-zero instead of degrading.

export ANCHORMD_LICENSE=ANMD-XXXX-XXXX-XXXX
export ANCHORMD_STRICT=1

Recommended for CI jobs, release pipelines, and any unattended run where silent unlicensed fallback is worse than a hard fail. See docs/strict.md for CI recipes (GitHub Actions, GitLab, CircleCI).

Framework Presets

Preset	Description
python-fastapi	FastAPI + async patterns
python-cli	Python CLI with typer/click
react-typescript	React + TypeScript + hooks
nextjs	Next.js App Router conventions
django	Django with ORM patterns
rust	Rust with clippy + proper error handling
go	Go with standard project layout
node-express	Express.js backend

Audit Scoring

Forge scores your CLAUDE.md on:

• Section coverage — does it have the essentials?
• Accuracy — does it match your actual codebase?
• Specificity — are instructions actionable or vague?
• Anti-patterns — does it prevent common mistakes?
• Freshness — is it up to date?

Development

# Clone and install
git clone https://github.com/Arete-Consortium/anchormd.git
cd anchormd
python3 -m venv .venv && source .venv/bin/activate
pip install -e ".[dev]"

# Run tests
pytest tests/ -v

# Lint
ruff check src/ tests/
ruff format src/ tests/

Community

Discord — Join the community

License

Business Source License 1.1 (BSL-1.1). See LICENSE for terms and the MIT change date.