claude-rails

agent
Guvenlik Denetimi
Uyari
Health Uyari
  • License — License: NOASSERTION
  • Description — Repository has a description
  • Active repo — Last push 0 days ago
  • Low visibility — Only 5 GitHub stars
Code Gecti
  • Code scan — Scanned 12 files during light audit, no dangerous patterns found
Permissions Gecti
  • Permissions — No dangerous permissions requested
Purpose
This is a file-based framework designed to add persistent memory, safety guardrails, and project structure to Claude Code. It uses cascading configuration files and hooks to prevent the AI from forgetting context between sessions and accidentally modifying files outside of designated project boundaries.

Security Assessment
The tool explicitly modifies your local system to function. It writes to the `~/.claude/` directory, creates global files like `~/CLAUDE.md`, and alters your `settings.json` to install execution hooks. While the automated code scan found no dangerous hidden patterns, hardcoded secrets, or outbound network requests, its core operation relies heavily on executing shell commands and managing local permissions. The developers are transparent about these risks and recommend using a dedicated machine, a virtual machine, or a fresh user account. Because it fundamentally alters your local development environment, the overall risk is rated as Medium.

Quality Assessment
The project is very new and currently labeled as a work-in-progress, meaning its structure and features may change rapidly. It has a low community visibility with only 5 GitHub stars, and the repository license is marked as NOASSERTION, which is a drawback for professional or commercial use. However, it appears to be under active maintenance and is thoroughly documented, featuring multiple languages and dedicated pages explaining its architecture.

Verdict
Use with caution: the framework is well-documented and safe from malicious code, but you should isolate it in a dedicated virtual machine or testing environment to prevent it from altering your primary system setup.
SUMMARY

Convention over Configuration for Claude Code. Guardrails instead of chaos. CLAUDE.md hierarchy, hooks, templates, examples — for Linux/Raspberry Pi.

README.md

🇩🇪 Deutsch · 🇬🇧 English · 🌐 Repo root (this page)

🚧 Work in Progress. This framework is actively developed. Structure, docs and features may change. Feedback welcome via Issues.

claude-rails

Stop fighting Claude. Start shipping with it.

Claude without the chaos — the setup serious Claude users
wish they had from day one.

A file-based framework that gives Claude Code persistent memory, project structure, safety guardrails and extensible skills — so it works with your projects instead of forgetting everything between sessions.

For anyone who juggles multiple projects in parallel and wants Claude to actually keep up.

⚠️ Use at your own risk. This framework writes into ~/.claude/, creates ~/CLAUDE.md, installs hooks and modifies settings.json. Backups are made but no rollback guarantee. Recommended: dedicated Pi or fresh test user account.

⚠️ Claude can hallucinate. Always verify critical outputs. See EN/docs/11-trust-but-verify.md.


✨ Features

Feature What it does
📁 CLAUDE.md Cascading Project context survives across sessions — global → project → sub-project, automatically loaded
📐 Master Template 7 mandatory sections per project. Convention over configuration — Claude knows where everything goes
🛡 Project Boundaries PreToolUse hooks enforce write isolation. Sessions can't mess with other projects
Skills Extensible slash commands (/crnote, /crdiscard, /crdiary ...) — add your own
🗑 Disposable Sessions Mark & auto-delete throwaway sessions. Keep your --resume list clean
📝 Central Scratchpad Cross-project notes via /crnote — ideas never get lost
🔍 Fresh-Eyes Review Spawn a zero-context agent to catch what a loaded session can't see
🚦 Houston Mode Safety switch: "Houston on" → analyze only. "go" → execute. One word to prevent disasters
🧹 Hat Counter Fun bet tracking between you and Claude. Because guardrails should have personality

💡 Why a sandbox?

Claude Code in full autonomy needs an isolated environment — don't risk your main machine. Pick what fits:

Option Cost Notes
Raspberry Pi 4/5 ~80 € Dedicated hardware, our reference setup. Setup guide
VM (Proxmox, VirtualBox, UTM) Free Any spare machine or your Mac/PC
Docker Free Synology NAS, any Linux host
WSL2 Free Windows — runs Linux natively
Fresh Linux user Free adduser claude-sandbox on an existing box

The framework uses chattr +a (Linux kernel feature) for INBOX hardening. Works out of the box on any Linux. macOS/Windows without WSL needs workarounds.


⚡ Real-world wins

  • 🚀 Tender response 4× faster — Upload price list + datasheets, Claude fills your tender in Excel format. Was 4h, now 1h. Sample template
  • 🧠 Knowledge base that researches itself — Dump meeting notes, Claude sorts them into the right folders. Workflows
  • 🛡 Safe sandbox on a 80 € Pi — Full autonomy without risking your main machine. Setup

📚 Examples

🏢 Work/ — Stakeholder-centric knowledge base. Sales, consulting, project management.

🏠 Home_Automation/ — PV, heat pump, sensors, automation flows.

🌐 HomeNet/ — Tiny 5-file project. Proves the framework scales down.

📔 Tagebuch/ — Diary/blog. See it live at trazer.online.


🛠 Skills

EN Command DE Command Purpose
/crnote /crnotiz Append to central scratchpad (or show list)
/crdiscard /crwegwerfen Mark session for deletion on exit
/crkeep /crbehalten Undo /crdiscard
/crdiary /crtagebuch Create a bilingual blog entry from session context
/crhatreport /crbesenreport Show hat counter score and history

Skills live in ~/.claude/skills/cr-*/SKILL.md — drop your own in, they just work.


🏗 Architecture at a Glance

~/CLAUDE.md                          Global rules (style, keywords, project overview)
  ↓ cascading
~/Project/CLAUDE.md                  Project rules (7 sections, imports master template)
  ↓ @import
~/.claude/schablonen/master_template.md   Structure + maintenance rules (shared)
  ↓ enforcement
~/.claude/hooks/                     Boundary hooks, session capture, cleanup
  ↓ actions
~/.claude/skills/cr-*/               Slash commands (note, discard, diary, ...)

Details: EN/docs/02-architecture.md


🚀 Quickstart

git clone https://github.com/trazer-labs/claude-rails.git
cd claude-rails
bash install.sh          # asks DE or EN, handles the rest

Fresh Pi? Run bash EN/onboarding/bootstrap.sh first. Details: EN/PREREQUISITES.md.

📦 What the installer does

  • Backs up everything before touching it (timestamped backup dir)
  • Diff-based skill updates — detects your local changes, never silently overwrites (creates .upstream for manual merge)
  • Auto-cleans old commands — migrates deprecated /commands/*.md to the new Skills system
  • Wires hooks into settings.json — boundary enforcement, session capture, cleanup
  • Sets up INBOX with chattr +a — kernel-level append-only protection

📖 Documentation

# Topic
1 Principles Core ideas on one page
2 Architecture Three layers, cascading, why pointers
3 Master Template The 7 sections explained
4 Maintenance Rules How the system stays consistent
5 Houston Mode The safety switch
6 Project Boundaries Write isolation + kernel hardening
7 Disposable Sessions Mark, delete, clean
8 Keywords Workflow "save", "done", "brainscan"
9 Consistency Check Superadmin structure audits
10 Scratchpad Cross-project notes
11 Trust but Verify The hallucination discipline
12 Fresh-Eyes Review Zero-context review pattern
13 Service Persistence systemd + registry pattern
14 Graphify Integration Knowledge graph for smarter navigation ✨ new

🇩🇪 Deutsche VersionDE/ · 🇬🇧 English versionEN/


🔧 Recommended Tools

Graphify — Knowledge graph generator (MIT, separate project). We run it across all our projects. One rule in the master template makes agents read the graph report before grepping blindly — result is fewer tokens and better answers on impact/dependency questions. Details: EN/docs/14-graphify-integration.md.


☕ Support this project

Ko-fi

⭐ A star on this repo helps too — it tells others this is worth looking at.


📜 License

CC-BY-SA 4.0 — do whatever you want, name Trazer as the source, share changes under the same license. See LICENSE.


Copyright © 2026 Trazer · CC-BY-SA 4.0

Yorumlar (0)

Sonuc bulunamadi