
spookyparticles
io.github.banana-intelligence-dot-ai/spookyparticles
Checks Vercel deploys, sites and cron jobs; sends the cause of a break to your coding agent to fix.
Documentation
Spooky Particles
VibeOps for apps on Vercel. You ship the app; Spooky keeps it running.
Live at spookyparticles.com. Sign in with Vercel.
What it does
- Watches schedules and deploys. Every Vercel cron job and schedule, with an alert when a run misses or fails and again when it recovers. A failed deploy comes with the log line that names the cause.
- Checks the basics. Uptime, server-error spikes, secrets in browser-readable variables, open preview links, unguarded cron routes, dependency updates, dates that expire. The full list is the Standard.
- Lists what you run. Every outside service each project depends on, what it costs on your Vercel bill, and whether it is still used.
- Hands work to your agent. An MCP server and a REST API, so a coding agent can ask what is broken and open a pull request to fix it. Setup for agents: spookyparticles.com/setup.md.
Skills
One agent skill per item of the Standard. Any coding agent that reads SKILL.md files (Claude Code, Codex, Cursor and others) can run one to bring a Vercel project up to that item. Each skill says how to check the item, how to fix it, and how to prove it is done. Every skill ends in a pull request or in steps for a person. None of them touches a secret's value or deletes anything. They work without Spooky; with Spooky, each names the API field or MCP tool that shows the item.
Install all of them for your agents:
npx skills add banana-intelligence-dot-ai/spookyparticles
Or copy one folder from skills/ into your agent's skills folder, for example .claude/skills/ for Claude Code.
Claude Code plugin
The plugin bundles every skill here plus the Spooky MCP server:
claude plugin marketplace add banana-intelligence-dot-ai/spookyparticles
claude plugin install spooky-particles@banana-intelligence
Or, on Claude Code 2.1.275 or later, in one step:
/plugin install spooky-particles --marketplace banana-intelligence-dot-ai/spookyparticles
The MCP server reads your key from the SPOOKY_API_KEY environment variable. Make a key at spookyparticles.com/app/connections and export it in your shell yourself. Never paste it into chat. Privacy: spookyparticles.com/privacy.
| # | Standard item | Skill | Spooky shows it in |
|---|---|---|---|
| 1 | Someone hears within 10 minutes when the live site stops answering | spooky-uptime | uptime.state in GET /api/v1/projects |
| 2 | Someone hears when a production deploy fails | spooky-deploy-alerts | whats_broken, GET /api/v1/incidents |
| 3 | Someone hears when server errors spike | spooky-error-alerts | reason needs_read_token in list_findings |
| 4 | Every scheduled job is watched | spooky-checkin | list_jobs, GET /api/v1/checkins |
| 5 | No secret sits in a variable the browser can read | spooky-public-secret | check secret_in_public_code in list_findings |
| 6 | Preview links are not open to the world | spooky-previews | check previews_open in list_findings |
| 7 | Scheduled routes refuse callers without the cron secret | spooky-cron-guard | check cron_unguarded in list_findings |
| 8 | Secrets are marked sensitive | spooky-sensitive | check secret_not_sensitive in list_findings |
| 9 | Dependencies get update pull requests on their own | spooky-deps | check deps_not_updated in list_findings |
| 10 | Nothing breaks on a date by surprise | spooky-expiry | list_expirations, GET /api/v1/expirations |
| 11 | Every outside service is listed, with cost and use | spooky-service | services[].usage, monthlyCostUsd in GET /api/v1/services |
| 12 | A spend limit is set | spooky-spend-limit | not checked yet; bill in GET /api/v1/services helps pick the amount |
The spooky skill (skills/spooky, also served at spookyparticles.com/skill.md) works through Spooky's findings one at a time and points to the skill for each item.
MCP server
Remote, streamable HTTP at https://spookyparticles.com/api/mcp, with Authorization: Bearer <your key>. Make a key at spookyparticles.com/app/connections. A read key runs the read tools; a write key adds the write tools. Nothing deletes. Listed in the official MCP Registry as io.github.banana-intelligence-dot-ai/spookyparticles.
claude mcp add --transport http spooky https://spookyparticles.com/api/mcp --header "Authorization: Bearer $SPOOKY_API_KEY"
Full tool list and setup for agents: spookyparticles.com/setup.md.
How it is built
Next.js (App Router) on Vercel, Neon Postgres through Drizzle, Better Auth, Vercel Workflow for long syncs. Every capability is an /api/v1 endpoint first; the web app, the MCP server and agents all use the same API.
apps/web: the app, the API and the MCP server. Run it locally with apps/web/README.md.packages/tier0: rules that sort billing and renewal email by sender, with no model call.docs/standard: what "done" means for a project on Vercel.skills: one agent skill per Standard item.
Security rules, enforced in code
- Read-only. Calls to your providers are reads, and nothing on your accounts is changed or deleted.
- Environment variable values are never stored. When Vercel returns one, it is dropped in memory.
- Every action is written to an append-only audit log.
- Revoking a connection deletes its tokens at once and, within 30 days, the data only that connection supplied.
Found a security problem? Email hello@spookyparticles.com rather than opening an issue.
License
MIT. See LICENSE.
No installation packages available.