Claude Code AI plugin. Mac automation, Telegram remote control, persistent memory with knowledge graph, heartbeat monitoring, research workflows, composable pipelines.
SuperClaw extends Claude Code with agentic Mac control, proactive system monitoring, and persistent memory. Remote control via Telegram, automated workflows via pipelines, academic research tools, and self-improving skills through dynamic code generation.
Key Capabilities:
- 39 specialized agents with 3-tier model routing (haiku/sonnet/opus)
- 13 deep skills (208-454 lines each, auto-generated via skill-forge)
- 4 slash commands for instant access
- 31 MCP tools across 3 servers (bridge, peekaboo, memory)
- 9 lifecycle hooks for keyword detection, tool enforcement, verification
- Persistent SQLite memory with FTS5 full-text search and knowledge graph
- Persistent cross-session memory with knowledge graph
- Screenshot capture (native + Peekaboo v3)
- UI element detection and interaction
- Window/app management
- AppleScript execution
- Click/type/scroll automation
- Send messages to phone from Claude
- Receive commands from Telegram
- Direct Telegram Bot API integration via TelegramPoller
- SQLite database with FTS5 full-text search
- Knowledge graph with entity/relationship tracking
- Conversation logging with semantic search
- Persistent memory with FTS5 search and knowledge graph
- System health checks (CPU, memory, disk)
- Process monitoring (dev servers, daemons)
- GitHub PR/issue tracking
- Calendar integration
- Sentry error tracking
- Custom metric collectors
- Proactive alerting via Telegram
- ArXiv paper fetching and analysis
- Literature review synthesis
- Experiment tracking and logging
- Statistical analysis
- Code review for research repositories
- Pre-built workflows (morning-brief, meeting-prep, deploy-monitor)
- Trigger-based automation
- Multi-step agent coordination
- Cron integration
- PR review summaries
- Unread notification digests
- Code ownership analysis
- Catch-up workflows after time away
- Node.js: 22+
- macOS (for Peekaboo Mac automation; memory/pipeline features work on any OS)
Everything else is auto-installed by the setup wizard.
# 1. Clone
git clone https://github.com/jaminitachi/SuperClaw.git ~/superclaw
cd ~/superclaw
# 2. One-command setup (installs everything, configures Telegram, registers plugin permanently)
npm run setup
# 3. Done! Just launch Claude Code — SuperClaw loads automatically
claudeThat's it. The setup wizard handles everything:
- Installs Peekaboo via Homebrew (Mac automation)
- Installs SoX (audio processing for TTS)
- Installs Node.js dependencies and builds 3 MCP servers
- Asks for Telegram bot token + chat ID (optional, interactive)
- Creates
superclaw.jsonwith all settings - Creates data directories
- Injects delegation rules into
~/.claude/CLAUDE.md - Registers SuperClaw as a permanent Claude Code plugin (no
--plugin-dirneeded)
If npm run setup fails to auto-register the plugin, or if you prefer manual installation:
# Register the local directory as a plugin marketplace
claude plugin marketplace add ~/superclaw
# Install the plugin permanently (user-scoped)
claude plugin install superclaw@superclawOnce registered, SuperClaw loads automatically every time you run claude — no --plugin-dir flag needed.
To verify registration:
claude plugin list # Should show superclaw@superclaw
claude plugin marketplace list # Should show ~/superclawTo uninstall:
claude plugin uninstall superclaw@superclaw
claude plugin marketplace remove superclaw[Telegram Bot API] ←→ [SuperClaw Bridge (TelegramPoller)]
↓
[Claude Code Plugin]
├── MCP Servers (3)
├── Skills, Agents, Hooks
└── SQLite Memory
↓
[Mac Control Layer]
├── Peekaboo
└── osascript
Data Flow:
- User says "send to phone: meeting at 3pm"
- UserPromptSubmit hook detects "send to phone" keyword
- telegram-control skill activates
- Delegates to mac-control agent (sonnet tier)
- Agent calls sc-bridge MCP tool
sc_telegram_send - Bridge sends message via Telegram Bot API
- Message delivered to Telegram
- Memory system logs interaction
| Domain | Agents | Models | Use Cases |
|---|---|---|---|
| Mac Automation | mac-control, mac-control-low | sonnet, haiku | Screenshots, UI automation, app control |
| Memory/Knowledge | memory-curator, memory-curator-low, memory-curator-high | sonnet, haiku, opus | Knowledge graph updates, semantic search, relationship extraction |
| Pipelines | pipeline-builder, pipeline-builder-high, workflow-monitor | sonnet, opus, haiku | Workflow design, multi-step automation, monitoring |
| System Health | heartbeat-mgr, system-monitor, system-monitor-high, cron-mgr | haiku, sonnet, opus | Proactive monitoring, alerting, cron management |
| Telegram | telegram-debugger | sonnet | Telegram Bot API diagnostics |
| Research | paper-reader, literature-reviewer, experiment-tracker, research-assistant, research-code-reviewer | sonnet, opus, haiku | Academic workflows, data analysis, experiment logging |
| Infrastructure | data-analyst, sc-verifier, setup-validator | sonnet, haiku | Data analysis, verification, installation validation |
| Developer Tools | sc-architect, sc-architect-low, sc-frontend, sc-code-reviewer, sc-code-reviewer-low, sc-debugger, sc-debugger-high, sc-test-engineer, sc-security-reviewer, sc-security-reviewer-low, sc-performance, sc-performance-high | opus, sonnet, haiku | Architecture analysis, frontend design, code review, debugging, testing, security audit, performance profiling |
| Orchestration | sc-prometheus, sc-metis, sc-momus, sc-atlas, sc-junior | opus, sonnet | 3-tier planning/execution/worker system for ultrawork mode |
| Meta | skill-forger | sonnet | Skill generation, dynamic code creation |
| Skill | Trigger | Description | Lines |
|---|---|---|---|
telegram-control |
"send to phone" | Send Telegram messages | 208 |
mac-control |
"screenshot", "click on" | Mac UI automation | 312 |
memory-mgr |
"remember this", "search memory" | Persistent knowledge operations | 287 |
heartbeat |
"system health", "check status" | Proactive monitoring | 341 |
automation-pipeline |
"pipeline", "morning brief" | Composable workflows | 389 |
cron-mgr |
"schedule", "every morning" | Cron job management | 234 |
setup |
"setup superclaw" | Installation wizard | 412 |
skill-forge |
"create skill" | Auto-generate new skills | 454 |
paper-review |
"read paper", "arxiv" | Academic paper analysis | 298 |
experiment-log |
"log experiment" | Research experiment tracking | 223 |
lit-review |
"literature review" | Multi-paper synthesis | 367 |
research-analysis |
"analyze data" | Statistical analysis | 276 |
dev-workflow |
"check PRs", "catch me up" | Developer productivity | 319 |
All skills are invoked via /superclaw:skill-name or auto-detected via keyword matching.
| Command | Description | Example |
|---|---|---|
/sc-status |
System health dashboard | Shows Telegram, memory, heartbeat status |
/sc-screenshot |
Quick screenshot via Peekaboo | Captures screen and analyzes content |
/sc-memory <query> |
Search memory database | /sc-memory "API design decisions" |
/sc-heartbeat |
Run health check now | Executes all collectors and reports |
sc_status- Check SuperClaw system statussc_telegram_send- Send message to Telegramsc_telegram_inbox- Fetch pending Telegram messagessc_telegram_status- Check Telegram Bot API connectionlist_cron_jobs- View scheduled tasksadd_cron_job- Schedule new jobremove_cron_job- Delete jobvalidate_cron_expression- Check cron syntax
screenshot- Capture screendetect_ui_elements- Find clickable elementsclick_element- Click by label/roletype_text- Keyboard inputlist_apps- Running applicationslist_windows- Open windowsfocus_window- Activate windowrun_applescript- Execute AppleScriptget_app_info- App metadataexecute_shortcut- Keyboard shortcutsscroll- Scroll in directiondrag- Drag and dropget_screen_bounds- Display dimensionswait_for_element- Element pollingget_element_text- Extract text from UI
store_knowledge- Save to memorysearch_knowledge- FTS5 full-text searchget_related_knowledge- Knowledge graph traversalupdate_knowledge- Modify entriesdelete_knowledge- Remove entrieslog_conversation- Save conversation turnsearch_conversations- Search logsget_memory_stats- Database statistics
Config file: ~/superclaw/superclaw.json
{
"telegram": {
"botToken": "your-bot-token",
"allowFrom": ["your-chat-id"],
"defaultChatId": "your-chat-id"
},
"peekaboo": {
"binaryPath": "/opt/homebrew/bin/peekaboo"
},
"heartbeat": {
"interval": 300000,
"collectors": ["system", "process", "github", "calendar"]
},
"memory": {
"dbPath": "~/superclaw/memory.db"
}
}Created automatically by /superclaw:setup.
- Telegram:
allowFromwhitelist for authorized chat IDs, bot token authentication - Config: 0600 permissions on token files
- Shell Safety: No user input in shell commands, all paths validated
- MCP Isolation: Each server runs in separate process
# Install dependencies
npm install
# Build plugin
npm run build
# Watch mode
npm run build:watch
# Run tests
npm test
# Type check
npm run typecheckTelegram Bot API connection failed:
# Verify bot token in config
cat ~/superclaw/superclaw.json | grep botToken
# Test bot token manually
curl https://api.telegram.org/bot<YOUR_TOKEN>/getMePeekaboo not found:
# Install via Homebrew
brew install peekaboo
# Or specify custom path in configMemory database locked:
# Check for stale connections
lsof ~/superclaw/memory.db
# Reset if needed
rm ~/superclaw/memory.db
/superclaw:setup # Re-run setupMIT
2.0.0