A complete Claude Code workflow cabin — memory, writing, publishing, remote control, and more.
Built through real-world AI collaboration, not theory. Every hook, every rule, every module earned its place by solving an actual problem.
┌─────────────────────┐
│ cc-cabin │
│ hooks + rules + │
│ workflow templates │
└──────────┬──────────┘
│
┌────────────┬───────────┼───────────┬────────────┐
│ │ │ │ │
┌─────┴─────┐ ┌────┴────┐ ┌───┴───┐ ┌────┴────┐ ┌─────┴─────┐
│ Memory │ │ Content │ │Remote │ │ Multi │ │ Clone │
│ │ │ │ │Control│ │ Session │ │ │
│ recallnest│ │alchemy +│ │ TG │ │ studio │ │ digital │
│ │ │publisher│ │bridges│ │ │ │ clone │
└───────────┘ └─────────┘ └───────┘ └─────────┘ └───────────┘
| Module | Description | Category |
|---|---|---|
| recallnest | Persistent memory MCP plugin. Local-first, never forgets. | Core |
| content-alchemy | 5-stage article pipeline — raw idea to polished draft | Content |
| content-publisher | Image generation, layout, and WeChat publishing | Content |
| digital-clone-skill | Create digital clones from corpus data | Content |
| telegram-ai-bridge | Command Claude Code from anywhere via Telegram | Remote |
| telegram-cli-bridge | Async CLI bridge for CC, Codex, and Gemini | Remote |
| claude-code-studio | Multi-session collaboration studio | Orchestration |
| openclaw-tunnel | Docker-to-host bridge for chat app integration | Orchestration |
All modules are optional. Install what you need.
git clone https://github.com/AliceLJY/cc-cabin.git
cd cc-cabin
bash install.shThe installer will:
- Check prerequisites (Claude Code, git, etc.)
- Show available modules — pick what you need
- Install hooks into
~/.claude/hooks/ - Set up CLAUDE.md and workflow templates
- Clone selected module repos
Battle-tested hooks that guard your workflow:
| Hook | Trigger | What it does |
|---|---|---|
bash-guard.sh |
PreToolUse:Bash | Blocks rm, enforces tmux for dev servers, PR guards |
edit-guard.sh |
PostToolUse:Edit | README bilingual check, API signature change alerts |
secret-guard.sh |
UserPromptSubmit | Detects API keys/secrets before they leak |
auto-commit.sh |
Stop | Reminds about uncommitted changes at session end |
session-summary.sh |
Stop | Auto-generates session index for later lookup |
track-edit.sh |
PostToolUse:Edit | Tracks which repos were touched per session |
nag-reminder.sh |
PostToolUse | ReAct drift detection — nudges observation notes |
pre-compact.sh |
PreCompact | Saves active state before context compression |
rating-capture.sh |
UserPromptSubmit | Captures 1-10 ratings for quality tracking |
ts-check.sh |
PostToolUse:Edit | Auto TypeScript syntax check after edits |
load-context.sh |
SessionStart | Injects identity and project status |
repos.conf |
— | Shared tracked repo list |
All hooks support profile control (CC_HOOK_PROFILE=off to disable all) and individual disable (CC_DISABLED_HOOKS=hook1.sh,hook2.sh).
- CLAUDE.md.example — Project contract template (execution rules, verification, reply discipline)
- workflow.md — Research → Plan → Implement methodology with ReAct loop
- memory/ — Memory index structure (MEMORY.md + identity + feedback templates)
The workflow methodology is opinionated:
- Three-stage discipline: Research → Plan → Implement. No skipping.
- ReAct loop: Act → Observe → Reflect → Record. Every step.
- Anti-sycophancy: Use neutral prompts. Don't preset conclusions.
- Task contracts: Define completion criteria before starting.
- Session isolation: Different tasks get different sessions.
| Variable | Default | Purpose |
|---|---|---|
CC_HOOK_PROFILE |
standard |
Set to off to disable all hooks |
CC_DISABLED_HOOKS |
— | Comma-separated list of hooks to skip |
CC_MEMORY_DIR |
auto-detect | Override project memory directory |
CC_GITHUB_ORGS |
— | Your GitHub org names (comma-separated), for PR guard |
CC_NAG_THRESHOLD |
5 |
Tool calls before ReAct reminder triggers |
Edit ~/.claude/hooks/repos.conf to add your project directories:
TRACKED_REPOS=(
"$HOME/Projects/my-app"
"$HOME/Projects/my-lib"
)This project was built by a non-developer through AI collaboration. The perspective is different from most CC workflow tools:
- Cross-domain, not just coding — Writing pipelines, bot operations, remote control, not just TDD
- Production-tested rules — Every rule in CLAUDE.md exists because its absence caused a real problem
- Memory as infrastructure — Persistent memory isn't optional; it's how the AI gets better over time
- Honest methodology — Anti-sycophancy prompts, neutral verification, confidence tags
See docs/architecture.md for how modules work together.
Just need memory + remote control? agent-nexus is a one-command installer for RecallNest + Telegram bridge — up and running in 60 seconds. cc-cabin is the full workflow scaffold; agent-nexus is the quick-start for the two most popular modules.
Issues and PRs welcome. If you've built hooks or workflow rules that made your CC life better, share them.
MIT