English | 한국어
A workspace operating system starter kit for Claude Code.
Turn Claude Code from a coding assistant into a general-purpose workspace OS.
Quick Start • Commands • Agents • Skills • Memory • Docs
git clone https://github.com/fivetaku/declaw.git
cd declawclaude/onboarding
This walks you through:
- reading the workspace structure
- filling in your identity and user contracts
- validating the scaffold
- choosing which capability packs to activate
/daily-note # start today's log
/todo fix auth bug # capture a task
/thinking-partner # structure a problem
/setup-workspace # validate scaffold anytime
When you clone declaw, you get a pre-structured workspace — not just a CLAUDE.md. It includes:
declaw/
├── CLAUDE.md # root constitution
├── INDEX.md # workspace navigation map
├── .claude/
│ ├── CLAUDE.md # control plane (imports contracts + rules)
│ ├── commands/ # 12 operator commands (slash commands)
│ ├── agents/ # 4 role-oriented agents
│ ├── rules/ # 7 modular rule files (auto-loaded)
│ ├── skills/ # core OS skills
│ ├── bootstrap/ # session start/setup/heartbeat checklists
│ └── settings.json # lifecycle hooks seed
├── 00-inbox/ # raw incoming capture
├── 00-system/
│ ├── 01-templates/ # daily note, todo, weekly review templates
│ ├── 02-contracts/ # IDENTITY, SOUL, USER, TOOLS
│ └── 04-docs/ # MEMORY_LANES, OPERATING_MODEL
├── 10-backlog/ # captured ideas and queued work
├── 20-testing/ # validation, gates, reviewer scenarios
├── 30-docs/ # user-facing guides and pack docs
├── 40-personal/
│ ├── 41-daily/ # human-facing daily logs
│ ├── 42-weekly/ # weekly reviews
│ └── 46-todos/ # active todos
├── 50-resources/ # supporting references
├── 90-archive/ # completed/retired material
├── memory/
│ ├── MEMORY.md # durable long-term memory
│ └── daily/ # agent continuity handoffs
Claude Code auto-loads CLAUDE.md, .claude/CLAUDE.md, and .claude/rules/*.md on every session. The contracts, bootstrap checklists, and memory are pulled in through @import and lifecycle hooks. You don't need to configure anything — just clone and start.
- Not just CLAUDE.md — A full workspace structure with contracts, memory lanes, commands, agents, skills, and hooks
- Not just a coding assistant — Supports research, operations, knowledge work, personal assistance, content pipelines, and code work
- Inspired by document-orchestrated agent workspaces — Rebuilt on Claude Code native features without depending on an external gateway product
- Repeatable sessions — Memory lanes + daily handoffs mean each session picks up where the last one left off
- Operator surface — 12 slash commands give you a workspace UI, not just a chat interface
- Role agents — Delegate to architect, reviewer, researcher, or operator with scoped tool access
- Extensible — Add your own commands, agents, skills, and capability packs on top of the core
Slash commands are the operator-facing workspace UI.
| Command | Description |
|---|---|
/onboarding |
First-run walkthrough — read structure, fill contracts, validate scaffold |
/setup-workspace |
Validate and complete the workspace scaffold |
/daily-note |
Open or create today's daily note |
/daily-review |
Review the day, update memory, write handoff |
/session-closeout |
Capture end-of-session state and write a clean handoff |
/weekly-review |
Create a weekly review from daily notes |
/todo [task] |
Capture a task quickly |
/todos |
Inspect current todo state |
/idea [idea] |
Capture an idea to inbox |
/thinking-partner |
Structure a problem through dialogue |
/inbox-processor |
Triage and route inbox items |
/sync |
Summarize and sync workspace changes |
/create-command |
Create a new operator command |
Role-oriented agents for delegated reasoning. These are long-lived role contracts, not one-off subagent calls.
| Agent | Role | Best for |
|---|---|---|
architect |
Structure, decomposition, system judgment | Design decisions, task breakdown, architecture review |
reviewer |
Risks, regressions, verification | Code review, quality gates, regression checks |
researcher |
Evidence gathering and synthesis | Documentation research, competitive analysis, fact-finding |
operator |
Workspace maintenance and organization | File management, cleanup, routine workspace operations |
Skills are reusable capabilities. Core skills power the workspace OS. Pack skills add domain-specific features.
| Skill | Description |
|---|---|
workspace-bootstrap |
Scaffold validation and initial setup |
memory-capture |
Capture decisions and knowledge into durable memory |
daily-sync |
Sync daily notes and continuity handoffs |
session-closeout |
End-of-session memory flush and handoff |
bridge-placeholder |
Extension point for external service integration |
onboarding |
First-run guided setup |
skill-builder |
Create new skills from repeated workflows |
| Pack | Skills | Description |
|---|---|---|
| Assistant | morning-briefing, email-automation, calendar, habit-tracker |
Personal assistant |
| Research | deep-research, web-monitor, data-analyst, transcribe |
Research and analysis |
| Content | content-draft, presentation, document-converter |
Content creation |
| Platform | apple-notes, apple-reminders |
macOS app integration |
| DevOps | docker-ops, ci-monitor |
Infrastructure monitoring |
declaw separates memory into three lanes to prevent drift and enable clean session handoffs.
| Lane | Location | Purpose | Audience |
|---|---|---|---|
| Durable | memory/MEMORY.md |
Stable knowledge, preferences, decisions | Agent (auto-loaded) |
| Continuity | memory/daily/YYYY-MM-DD.md |
Session handoff — what was done, what's next | Agent (session start) |
| Daily Log | 40-personal/41-daily/YYYY-MM-DD.md |
Human-readable work journal | Human |
See 00-system/04-docs/MEMORY_LANES.md for the full lane model.
declaw uses modular contracts instead of one giant CLAUDE.md.
| Contract | Claude Code mapping | Purpose |
|---|---|---|
IDENTITY.md |
.claude/rules/10-identity.md |
Who this workspace agent is |
SOUL.md |
.claude/rules/10-identity.md |
Core values and behavior principles |
USER.md |
CLAUDE.local.md |
Who uses this workspace |
TOOLS.md |
.claude/rules/40-tooling.md |
How tools should be used |
Full contracts live in 00-system/02-contracts/. Summaries are auto-loaded via .claude/rules/.
Hooks automate workspace discipline at session boundaries.
| Event | What it does |
|---|---|
SessionStart |
Injects boot checklist and recent memory context |
Stop |
Reminds to capture memory and write handoff |
PreCompact |
Prompts memory flush before context compression |
Setup:init |
Suggests /setup-workspace on first run |
SubagentStart |
Scoping reminder for delegated work |
SubagentStop |
Result format verification |
Configured in .claude/settings.json.
After onboarding, you do not have to stay in the terminal. declaw is designed to work through Remote Control, native messaging channels, and Scheduled Tasks.
When used via channels or remote, the same workspace contracts, commands, and skills still apply. The interaction surface changes, but the operating model stays the same.
Start a remote session:
claude remote-controlThis displays a session URL and QR code. Open the URL on any device or scan the QR with the Claude mobile app. Your local files, MCP servers, and workspace stay on your machine.
To enable for all sessions:
/config → Enable Remote Control for all sessions → true
Channels push external messages directly into an active Claude Code session.
Typical pattern:
- Enable a supported Claude Code channel using the current official workflow.
- Pair or authorize the sender.
- Route the incoming work through declaw commands, skills, and memory lanes.
Use .claude/skills/bridge-placeholder/SKILL.md as the canonical place to adapt channel and bridge setup to the current Claude Code release. Do not treat README examples as fixed CLI syntax.
| Method | Runs on | Machine required | Min interval |
|---|---|---|---|
/loop |
Your machine | Yes (session open) | 1 min |
| Cloud tasks | Claude-hosted scheduled execution | No | 1 hour |
| Desktop tasks | Your machine | Yes (app open) | 1 min |
In-session polling:
/loop 5m check project status and alert if issues
/loop 30m /daily-review
Cloud scheduled tasks (runs 24/7 without your machine):
/schedule daily morning briefing at 9am
Or configure at claude.ai/code/scheduled.
| Feature | Trigger | Best for |
|---|---|---|
| Remote Control | You drive from browser/phone | Steering in-progress work from another device |
| Channels | External event pushed in | React to messages on Discord/Telegram while away |
| Scheduled Tasks | Cron/interval | Morning briefings, periodic health checks, recurring reviews |
All three keep your files local and require claude.ai login.
- Not a gateway product — It composes Claude Code features into a workspace operating model
- Not only a coding assistant plugin — Coding is one use case among many
- Not only a skill collection — Skills sit on top of the workspace OS core
- Not a CLAUDE.md template — It's a full operating structure with 7 layers
| Document | Description |
|---|---|
INDEX.md |
Workspace navigation map |
30-docs/GETTING_STARTED.md |
First-run guide with examples |
30-docs/PACKS.md |
Capability pack overview |
00-system/04-docs/MEMORY_LANES.md |
Memory lane model and policy |
00-system/04-docs/OPERATING_MODEL.md |
End-to-end operating flow |
00-system/02-contracts/README.md |
Contract layer explanation |
- Claude Code CLI
- Claude Max/Pro subscription or a supported Claude API key
No other dependencies. No npm install. No build step. Just clone and start.
MIT
Claude Code is the only claw you need.