cybrix-deploy

Prerequisites

Before deploying, ensure the user has an API token. Check in this order:

1. CYBRIX_DEPLOY_TOKEN — set automatically by Claude Code from userConfig keychain (no action needed).
2. Environment variable CYBRIX_TOKEN.
3. File ~/.config/cybrix/token.
4. File .cybrix/token in the project (gitignored).

If none exist, instruct the user exactly like this:

No Cybrix token found. Get one free at https://app.cybrix.cc/dashboard (Step 3 — "Save your API token" → Generate).

Once you have it, you can either:

• Paste it here and I'll use it for this deploy (easiest)
• Run export CYBRIX_TOKEN=<token> in your terminal to set it for this session
• Run echo <token> > ~/.config/cybrix/token to save it permanently

Then wait for the user to provide the token. If they paste it directly in chat, use it immediately — do not require them to re-run any command. The project is ready to deploy once the token is available.

Deployment workflow

Step 1 — Detect project type (heuristic)

Do not rely on a framework whitelist. Instead, look for signals and classify the project as static, server, or unknown.

Static signals (proceed with deploy):

• package.json has a build script AND output lands in dist/, out/, build/, public/, _site/, or .output/public/
• next.config.{js,ts,mjs} with output: 'export' or output: 'static'
• astro.config.{js,ts,mjs} present (default mode is static)
• vite.config.{js,ts} without SSR plugins
• _config.yml (Jekyll), config.toml or hugo.toml (Hugo), .eleventy.js / eleventy.config.js (Eleventy), zola.toml (Zola)
• Only HTML/CSS/JS files at root, no server entry point

Server signals (refuse — see below):

• Dockerfile or docker-compose.yml (unless it only copies a static dist/)
• main.go, server.go, or any *.go containing net/http or ListenAndServe
• main.py, app.py, server.py with uvicorn, gunicorn, flask, fastapi imports, or a if __name__ == '__main__' block calling app.run / serve / asyncio.run
• main.rs or server.rs with actix, axum, rocket, warp, or tokio::main
• package.json with a start script that runs node/tsx/bun on a server file (NOT next start in a static-export config)
• Gemfile with puma, unicorn, rails, or sinatra
• pom.xml or build.gradle with spring-boot
• .csproj with ASP.NET

Database signals (warn but allow if everything else is static):

• *.sql files, migrations/ folder, prisma/schema.prisma, drizzle.config.*, DATABASE_URL referenced in source
• A static site hitting a hosted DB from the browser is unusual but valid. Warn the user, don't refuse.

When refusing (server signals detected):

This looks like a project that needs a server runtime — I detected <specific signal, e.g. "main.go with net/http" or "Dockerfile with EXPOSE">.

Cybrix currently supports static sites only. Your options:

1. Convert to a static export (e.g. Next.js output: 'export', Astro, Hugo).
2. Use a service that supports backends: Railway, Fly.io, Render.
3. Tell me to deploy anyway if you think the detection is wrong.

Always allow option 3 — heuristics are imperfect and the user knows their project.

Static output directories to check, in order: dist, out, public, _site, build, .output/public.

Step 2 — Scan environment variables

After confirming the project is static but BEFORE running the build, scan for environment variables the build will need.

2a. Read .env files — parse KEY=value format, skip comments (#) and blank lines. Files to check: .env, .env.local, .env.production, .env.example.

2b. Grep source code for build-time env var references:

• JS/TS: process.env.X, import.meta.env.X
• Look in src/, app/, pages/, components/ — any *.{js,jsx,ts,tsx,vue,svelte}
• Pay extra attention to NEXT_PUBLIC_*, VITE_*, PUBLIC_*, REACT_APP_* — these are baked into the bundle at build time

2c. Cross-reference keys found in code against keys present in .env files to find what the build needs.

2d. Show the user:

I detected the following environment variables your build needs:

NEXT_PUBLIC_API_URL (in .env.local, used in 3 files) VITE_STRIPE_KEY (in .env.local, used in src/checkout.ts)

These need to be set before the build. How would you like to provide them?

1. Paste them here (sent encrypted with the deploy)
2. Set them later in the dashboard
3. Skip (build may fail or site may not work correctly)

If the user picks option 1, ask for each value one at a time. Include them in the multipart POST to /v1/deploys as an env_vars field (JSON map: {"KEY": "value", ...}).

2e. Warn about missing variables — if a var is referenced in code but not in any .env file:

⚠ AUTH_SECRET is referenced in your code but not in any .env file. Provide it now or the build may fail.

2f. Refuse to forward secrets in client-exposed vars — if a key with a client prefix (NEXT_PUBLIC_*, VITE_*, REACT_APP_*, PUBLIC_*) looks like a secret (*_SECRET, *_PRIVATE_KEY, DATABASE_URL, JWT_SECRET):

⚠ NEXT_PUBLIC_JWT_SECRET looks like a private secret but has a client-bundle prefix — it will be visible to anyone who opens your site's source. Are you sure you want to include it?

Do not send it without explicit confirmation.

Step 3 — Choose project name and confirm

3a. Infer a default name from the current folder name, slugified (lowercase, hyphens, max 32 chars). Example: my-portfolio-site.

3b. Check availability by calling:

GET https://api.cybrix.cc/v1/slugs/<name>/available

Response: {"available": true, "slug": "my-portfolio-site"}

• If available: true — use it as the default.
• If available: false — do NOT use it. Tell the user:

  The name <name> is already taken. What would you like to call your project? It will be live at <your-name>.cbrx.cc. Then check availability of the new name too. Repeat until available.

3c. Present summary (only after confirming name is available):

Ready to deploy to Cybrix:

• Project name: → live at <name>.cbrx.cc
• Build command:
• Output directory:
• Env vars: included / none detected

Continue? (yes / change name / change build / change output)

If the user says change name, ask for a new name and re-check availability. Use the confirmed answers to override defaults before proceeding.

Step 4 — Build

Run the build command in the project root. Stream output to the user.

If the build fails, do not retry. Show the last 40 lines and say:

Build failed. Fix the error above and try again.

Step 5 — Deploy

Run ${CLAUDE_PLUGIN_ROOT}/scripts/deploy.sh <project_name> <output_dir>. The script:

1. Creates a project via POST /v1/projects with {"name": "<project_name>"}, receives { id, slug }. Skips if CYBRIX_PROJECT_ID is already set.
2. Tars and gzips the output directory.
3. POSTs the tarball to https://api.cybrix.cc/v1/deploys as multipart form data with fields project_id (UUID) and file (the .tar.gz), optionally env_vars (JSON map). Includes Authorization: Bearer $CYBRIX_TOKEN.
4. Receives { id, project_id, status } in the response.
5. Polls https://api.cybrix.cc/v1/deploys/<id> every 2 seconds until status is live or failed (max 5 minutes).
6. Prints the result as JSON on stdout including deployed_url and slug.

Step 6 — Report to the user

On success:

Deployed.

Live: https://.cbrx.cc Dashboard: https://app.cybrix.cc/dashboard/projects/<project_id>

On failure:

Deploy failed.

Reason: Logs: https://app.cybrix.cc/deployments/

Caching

After the first successful deploy, cache {project_id, slug} in .cybrix/project.json. On subsequent deploys, skip the "confirm project name" step unless the user explicitly asks to deploy a different project.

Also add .cybrix/ to .gitignore if it doesn't already contain it.

Errors

API status	Skill behavior
401	Token invalid/revoked. Tell user to refresh at app.cybrix.cc/dashboard.
402	Free tier project limit hit. Show upgrade link: cybrix.cc/pricing.
413	Tarball >100 MB. Suggest auditing output for large assets.
429	Wait 30s, retry once.
5xx	Show error, suggest retry, link to app.cybrix.cc/dashboard.

What this skill does NOT do

• Does not touch Git history.
• Does not push to or read from Git remotes.
• Does not modify project source.
• Does not store secrets anywhere except the user's machine.
• Does not deploy anything outside Cybrix.

If the user asks for any of these, redirect to the appropriate tool.