Deploy to Vercel

Deploy any project to Vercel. Always deploy as preview (not production) unless the user explicitly asks for production.

The goal is to get the user into the best long-term setup: their project linked to Vercel with git-push deploys. Every method below tries to move the user closer to that state.

Step 1: Gather Project State

Run all four checks before deciding which method to use:

CODEBLOCK0

Team selection

If the user belongs to multiple teams, present all available team slugs as a bulleted list and ask which one to deploy to. Once the user picks a team, proceed immediately to the next step — do not ask for additional confirmation.

Pass the team slug via --scope on all subsequent CLI commands (vercel deploy, vercel link, vercel inspect, etc.):

CODEBLOCK1

If the project is already linked (.vercel/project.json or .vercel/repo.json exists), the orgId in those files determines the team — no need to ask again. If there is only one team (or just a personal account), skip the prompt and use it directly.

About the .vercel/ directory: A linked project has either:

• - .vercel/project.json — created by vercel link (single project linking). Contains projectId and orgId.
• INLINECODE12 — created by vercel link --repo (repo-based linking). Contains orgId, remoteName, and a projects array mapping directories to Vercel project IDs.

Either file means the project is linked. Check for both.

Do NOT use vercel project inspect, vercel ls, or vercel link to detect state in an unlinked directory — without a .vercel/ config, they will interactively prompt (or with --yes, silently link as a side-effect). Only vercel whoami is safe to run anywhere.

Step 2: Choose a Deploy Method

Linked (.vercel/ exists) + has git remote → Git Push

This is the ideal state. The project is linked and has git integration.

1. 1. Ask the user before pushing. Never push without explicit approval:

CODEBLOCK2

1. 2. Commit and push:

   git add .
   git commit -m "deploy: <description of changes>"
   git push
   

Vercel automatically builds from the push. Non-production branches get preview deployments; the production branch (usually main) gets a production deployment.

1. 3. Retrieve the preview URL. If the CLI is authenticated:

   sleep 5
   vercel ls --format json
   

The JSON output has a deployments array. Find the latest entry — its url field is the preview URL.

If the CLI is not authenticated, tell the user to check the Vercel dashboard or the commit status checks on their git provider for the preview URL.

---

Linked (.vercel/ exists) + no git remote → vercel deploy

The project is linked but there's no git repo. Deploy directly with the CLI.

CODEBLOCK5

Use --no-wait so the CLI returns immediately with the deployment URL instead of blocking until the build finishes (builds can take a while). Then check on the deployment status with:

CODEBLOCK6

For production deploys (only if user explicitly asks):

vercel deploy [path] --prod -y --no-wait

---

Not linked + CLI is authenticated → Link first, then deploy

The CLI is working but the project isn't linked yet. This is the opportunity to get the user into the best state.

1. 1. Ask the user which team to deploy to. Present the team slugs from Step 1 as a bulleted list. If there's only one team (or just a personal account), skip this step.

1. 2. Once a team is selected, proceed directly to linking. Tell the user what will happen but do not ask for separate confirmation:

CODEBLOCK8

1. 3. If a git remote exists, use repo-based linking with the selected team scope:

   vercel link --repo --scope <team-slug>
   

This reads the git remote URL and matches it to existing Vercel projects that deploy from that repo. It creates .vercel/repo.json. This is much more reliable than vercel link (without --repo), which tries to match by directory name and often fails when the local folder and Vercel project are named differently.

If there is no git remote, fall back to standard linking:

   vercel link --scope <team-slug>
   

This prompts the user to select or create a project. It creates .vercel/project.json.

1. 4. Then deploy using the best available method:

- If a git remote exists → commit and push (see git push method above) - If no git remote → vercel deploy [path] -y --no-wait --scope <team-slug>, then vercel inspect <url> to check status

---

Not linked + CLI not authenticated → Install, auth, link, deploy

The Vercel CLI isn't set up at all.

1. 1. Install the CLI (if not already installed):

CODEBLOCK11

1. 2. Authenticate:

   vercel login
   

The user completes auth in their browser. If running in a non-interactive environment where login is not possible, skip to the no-auth fallback below.

1. 3. Ask which team to deploy to — present team slugs from vercel teams list --format json as a bulleted list. If only one team / personal account, skip. Once selected, proceed immediately.

1. 4. Link the project with the selected team scope (use --repo if a git remote exists, plain vercel link otherwise):

CODEBLOCK13

1. 5. Deploy using the best available method (git push if remote exists, otherwise vercel deploy -y --no-wait --scope <team-slug>, then vercel inspect <url> to check status).

---

No-Auth Fallback — claude.ai sandbox

When to use: Last resort when the CLI can't be installed or authenticated in the claude.ai sandbox. This requires no authentication — it returns a Preview URL (live site) and a Claim URL (transfer to your Vercel account).

CODEBLOCK14

Arguments:

• - path - Directory to deploy, or a .tgz file (defaults to current directory)

Examples:
CODEBLOCK15

The script auto-detects the framework from package.json, packages the project (excluding node_modules, .git, .env), uploads it, and waits for the build to complete.

Tell the user: "Your deployment is ready at [previewUrl]. Claim it at [claimUrl] to manage your deployment."

---

No-Auth Fallback — Codex sandbox

When to use: In the Codex sandbox where the CLI may not be authenticated. Codex runs in a sandboxed environment by default — try the CLI first, and fall back to the deploy script if auth fails.

1. 1. Check whether the Vercel CLI is installed (no escalation needed for this check):

CODEBLOCK16

1. 2. If vercel is installed, try deploying with the CLI:

CODEBLOCK17

1. 3. If vercel is not installed, or the CLI fails with "No existing credentials found", use the fallback script:

CODEBLOCK18

The script handles framework detection, packaging, and deployment. It waits for the build to complete and returns JSON with previewUrl and claimUrl.

Tell the user: "Your deployment is ready at [previewUrl]. Claim it at [claimUrl] to manage your deployment."

Escalated network access: Only escalate the actual deploy command if sandboxing blocks the network call (sandbox_permissions=require_escalated). Do not escalate the command -v vercel check.

---

Agent-Specific Notes

Claude Code / terminal-based agents

You have full shell access. Do NOT use the /mnt/skills/ path. Follow the decision flow above using the CLI directly.

For the no-auth fallback, run the deploy script from the skill's installed location:

bash ~/.claude/skills/deploy-to-vercel/resources/deploy.sh [path]

The path may vary depending on where the user installed the skill.

Sandboxed environments (claude.ai)

You likely cannot run vercel login or git push. Go directly to the no-auth fallback — claude.ai sandbox.

Codex

Codex runs in a sandbox. Check if the CLI is available first, then fall back to the deploy script. Go to the no-auth fallback — Codex sandbox.

---

Output

Always show the user the deployment URL.

• - Git push: Use vercel ls --format json to find the preview URL. If the CLI isn't authenticated, tell the user to check the Vercel dashboard or commit status checks.
• CLI deploy: Show the URL returned by vercel deploy --no-wait. Use vercel inspect <url> to check build status and report it to the user.
• No-auth fallback: Show both the preview URL and the claim URL:

CODEBLOCK20

Do not curl or fetch the deployed URL to verify it works. Just return the link.

---

Troubleshooting

Network Egress Error (claude.ai)

If deployment fails due to network restrictions on claude.ai, tell the user:

CODEBLOCK21

Escalated Network Access (Codex / sandboxed environments)

If deployment fails due to network issues (timeouts, DNS errors, connection resets) in a sandboxed environment, rerun with escalated permissions (sandbox_permissions=require_escalated). Do not escalate the initial CLI availability check — only escalate the actual deploy command.

Example guidance to the user:
CODEBLOCK22

CLI Auth Failure

If vercel login or vercel deploy fails with authentication errors, fall back to the no-auth deploy script (claude.ai or Codex variant, depending on the environment).