overclaud

Optimize how Claude works for you — across every surface.

Đang nâng cấp lên đa-agent → Hatch. overclaud (single-agent) đang được nâng thành một harness điều phối nhiều coding agent (Claude Code, Codex, Kiro, Antigravity) trên cùng một repo, tối ưu token, theo quy trình kiểu Agile. overclaud hiện tại trở thành compiler backend cho Claude bên trong Hatch. Sản phẩm thuộc hệ sinh thái Finolabs. Xem bộ thiết kế ở hatch/.

Claude reads your instructions every message, but most users either don't set them up, put everything in one place, or waste tokens on vague guidance. overclaud fixes that.

What This Is

A complete instruction optimization system for Claude:

Templates — Ready-to-use instructions for 5 roles and 6 project types
Layer architecture — Know exactly where to put each instruction
Token optimization — Every word earns its place
Installable skill — Let Claude set everything up for you

The Problem

Claude has 9+ surfaces where instructions can live. Each has different scope, precedence, and purpose:

Managed Policy     →  Org-wide (IT-deployed)
User CLAUDE.md     →  Personal, all projects     ← most users stop here
User Rules         →  Personal, path-scoped
Project CLAUDE.md  →  Team-shared, one project
Project Rules      →  Team-shared, path-scoped
Subdirectory       →  On-demand loading
Personal Prefs     →  Chat & Cowork settings
Cowork Instructions→  Per Cowork project
System Prompt      →  API / automation

Most users dump everything into one CLAUDE.md. The result: bloated context, wrong scope, wasted tokens every single message.

Quick Start

Option 1: Install as a Skill (Recommended)

Download overclaud.skill from this repo, then:

Claude Code:

claude install-skill overclaud.skill

Cowork: Drop the .skill file into your Cowork session.

Then tell Claude: "Set up my Claude instructions"

Claude will ask about your role, use cases, and preferences — then generate and install optimized instructions in the right places.

Option 2: Bootstrap Prompt (Any Surface)

Copy this into any Claude conversation:

Fetch and follow the setup guide at:
https://raw.githubusercontent.com/Fioenix/overclaud/main/skill/SKILL.md

Read the reference files in skill/references/ and templates in skill/templates/.
Ask me about my role and preferences, then help me set up optimized Claude instructions.

Option 3: Auto-Suggest After /init (Claude Code)

During setup, the skill offers to install a hook that auto-suggests optimization whenever you
run /init on a new project. Claude handles the installation — no manual config needed.

You: set up my Claude instructions
Claude: [runs overclaud skill, sets everything up]
Claude: "Want me to enable auto-optimization? Next time you /init a project,
         I'll suggest optimizing the generated CLAUDE.md automatically."
You: yes
Claude: ✓ Hook installed. You're all set.

Option 4: Browse and Copy

Explore the skill/templates/ directory and copy what fits:

Need	Browse
Global instructions by role	`templates/global/`
Project-level by use case	`templates/project/`
Cowork project instructions	`templates/cowork/`
Personal Preferences text	`templates/personal-preferences/`
Path-scoped rules	`templates/rules/`

Instruction Layer Map

Understanding where to put instructions is half the battle:

What to Configure	Where to Put It	Why There
Your role, communication style, language	`~/.claude/CLAUDE.md`	Applies everywhere, personal
Code style for a specific project	`.claude/rules/code-style.md`	Scoped to project, path-specific
Team coding standards	`./.claude/CLAUDE.md`	Shared via git
Your preference for deep-dive answers	Personal Preferences (Settings)	Applies in Chat & Cowork
Cowork daily ops workflow	Cowork Project Instructions	Scoped to one Cowork project

See skill/references/layers.md for the complete architecture.

Token Optimization

Your global CLAUDE.md loads into every message. At 100 messages/day:

Extra tokens	Daily waste	Monthly waste
50	5,000	150,000
200	20,000	600,000
500	50,000	1,500,000

overclaud templates are designed to maximize signal per token. See skill/references/token-optimization.md for techniques.

Repo Structure

overclaud/
├── README.md
├── skill/
│   ├── SKILL.md                    # Installable skill — the setup agent
│   ├── templates/
│   │   ├── global/                 # Global CLAUDE.md by role (5 templates)
│   │   ├── project/                # Project CLAUDE.md by use case (6 templates)
│   │   ├── cowork/                 # Cowork project instructions (3 templates)
│   │   ├── personal-preferences/   # Settings UI text (3 templates)
│   │   └── rules/                  # .claude/rules/ examples (3 templates)
│   └── references/
│       ├── layers.md               # Complete layer architecture
│       ├── token-optimization.md   # Token efficiency techniques
│       └── common-mistakes.md      # Anti-patterns to avoid
├── guides/
│   ├── layering-strategy.md        # What goes where — decision guide
│   └── contributing-templates.md   # How to contribute new templates
└── hatch/                          # Multi-agent upgrade (design docs)
    ├── README.md                   # Hatch — multi-agent harness overview
    ├── docs/                       # 00-vision … 08-roadmap
    └── spec/                       # registry / ticket / ledger schemas

Contributing

We welcome contributions! See CONTRIBUTING.md for guidelines.

Template contributions are especially welcome — the more roles and use cases covered, the more useful this becomes for the community.

License

MIT — use it however you want.