Superpowers

All installed tools that supercharge the development workflow. Custom skills, MCP servers, monitoring, and automation.

Quick Reference

Command / Tool	What it does	Type
`/deploy <slug>`	Build + deploy + restart + verify	Skill
`/vps-status [vps1\|vps2\|all]`	PM2 + disk + memory report	Skill
`/health-check [slug\|all]`	HTTP + PM2 health for all projects	Skill
`/db-migrate <slug>`	Backup + apply migrations	Skill
`/ops-register`	Register current project in ops-dashboard	Skill
`/singularity-create`	Create a skill with evolution tracking	Plugin
`/singularity-score`	Evaluate skill quality (5 dimensions)	Plugin
`/singularity-dashboard`	View skill maturity status	Plugin
`/loop 5m <prompt>`	Recurring task during session	Built-in
`/schedule`	Create cloud tasks (survives restarts)	Built-in
`SSH MCP tools`	ssh_execute, ssh_upload, etc.	MCP
`Prometheus MCP`	execute_query, list_metrics, etc.	MCP

Custom Skills

skills

5 custom skills available in every project. Located at ~/.claude/skills/.

/deploy — Reads projects.yaml, builds, SCPs, restarts PM2, verifies health
/vps-status — PM2 table, disk, memory, load averages; flags excessive usage
/health-check — Curls endpoints + PM2 SSH; reports against thresholds
/db-migrate — Backs up first, shows pending files, handles PostgreSQL + SQLite
/ops-register — Auto-detects project, generates YAML entry, rebuilds dashboard

/deploy estonian-accounting/vps-status all/health-check all/db-migrate estonian-accounting/ops-register

SSH MCP Server

mcp

37 tools for direct VPS management without writing SSH commands.

VPS1 — 37.27.34.70 (Hetzner, scraping/betting)
VPS2 — 46.62.129.138 (Zone.ee, apps)
ssh_execute — Run any command on a VPS
ssh_upload / ssh_download — Transfer files
ssh_system_health — CPU, memory, disk, load
ssh_database_query — Run DB queries remotely
ssh_backup_create — Create backups

ssh_execute(server="VPS2", command="pm2 jlist")ssh_system_health(server="VPS1")

Prometheus MCP

mcp

6 tools to query application metrics directly from Claude Code sessions.

execute_query — Run PromQL query
execute_range_query — Time-series query
list_metrics — Show all available metrics
get_metric_metadata — Describe a metric
get_targets — Show scrape targets
Requires Prometheus on VPS-2 port 9090

execute_query("up")list_metrics()

singularity-claude

plugin

5-stage skill evolution loop that tracks and improves skill quality over time.

Draft -> Tested (3+ runs, 60+ avg) -> Hardened (5+ runs, 80+ avg) -> Crystallized (90+)
5 scoring dimensions: Correctness, Completeness, Edge cases, Efficiency, Reusability
Data stored at ~/.claude/singularity/

/singularity-create/singularity-score/singularity-repair/singularity-crystallize/singularity-dashboard

HTTP Hooks -> Activity Feed

monitoring

Every Bash, Edit, and Write action automatically posts to the ops-dashboard activity feed.

PostToolUse hook fires for Bash/Edit/Write
HTTP POST to ops-dashboard /api/webhook
Dashboard parses event, detects project, generates summary
Configured in ~/.claude/settings.json under hooks.PostToolUse

Activity page: /activity

Agent Teams

agent

Multi-agent parallel coordination for complex cross-project work.

Enabled via CLAUDE_CODE_EXPERIMENTAL_AGENT_TEAMS=1
Claude creates a team with TeamCreate, spawns teammates
Shared task list prevents conflicts
3-5 teammates max, each owns different files
Use plan-approval mode for risky changes

TeamCreateSendMessageTaskCreateTaskUpdate

Jarvis AI Ops

monitoring

Framework for 24/7 headless Claude monitoring with 16 cron tasks + Discord alerting.

Every 10 min — PM2 process watchdog
Every 15 min — HTTP ping sweep (13 endpoints)
Every 30 min — VPS health, disk alerts
Daily — Morning standup, infra report, evening synthesis
Weekly — Security scan, 714+ tests, SSL check, log cleanup
Nexus CIG MCP for 98% context compression

Location: C:\Users\sysde\projects\jarvis

Mission Control

dashboard

Open-source dashboard for tracking all Claude Code sessions, costs, and tasks.

100+ Claude Code sessions auto-discovered
Token/cost tracking per session
101 API endpoints (Scalar UI at /docs)
Kanban task board + session replay
URL: http://localhost:3800

pm2 logs mission-controlhttp://localhost:3800

Cloud Scheduled Tasks

built-in

Run prompts on Anthropic cloud infrastructure -- works even when your machine is off.

Daily PR review across all repos
Weekly dependency audit
Nightly CI failure summaries
Min interval: 1 hour. Needs GitHub repos.
Visit: https://claude.ai/code/scheduled

/schedule

Architecture Overview

You (Claude Code session)
  |-- Custom Skills (/deploy, /vps-status, /health-check, /db-migrate, /ops-register)
  |-- SSH MCP -> VPS-1 (37.27.34.70) + VPS-2 (46.62.129.138)
  |-- Prometheus MCP -> Application metrics
  |-- HTTP Hooks -> Ops Dashboard activity feed
  |-- Agent Teams -> Parallel cross-project work
  +-- singularity-claude -> Skill evolution tracking

Ops Dashboard (https://ops-46-62-129-138.sslip.io/)
  |-- 30 projects registered
  |-- Health checks (15-min HTTP ping + nightly full SSH sweep)
  |-- Self-evolving engine (baselines, anomaly detection, action items)
  |-- Activity feed (from HTTP hooks)
  |-- Resources/docs per project
  +-- Weekly digest

Mission Control (http://localhost:3800)
  |-- 100+ Claude Code sessions tracked
  |-- Token/cost analytics
  +-- Kanban task board

Jarvis (pending Discord setup)
  |-- 16 cron tasks
  |-- 24/7 headless Claude monitoring
  +-- Discord alerting