Vercel Token Setup

What the source does

• FilesystemNone observedNo reads or writes in the source
• NetworkNone observedNo outbound calls in the source
• ShellNone observedNo commands run in the source
• DependenciesPinnedNothing floats
• SecretsNone foundNo keys in source

Five audit layers

Gates applied: no behavioural pass.

assistant/CLAUDE.md

assistant/src/approvals/CLAUDE.md

assistant/src/notifications/CLAUDE.md

assistant/src/runtime/CLAUDE.md

gateway/CLAUDE.md

Audit 0.4.1 · source sha 19bc86343b63 · 14 Oct 2026Report an issue or request a re-scan

• Publishervellum-ai GitHub organizationOwner of the repository this skill is read from
• Source foldervellum-ai/vellum-assistant/skills/vercel-token-setupThe exact folder we audited
• LicenceMITAs the repository states it
• Commit pinned for this verdict19bc86343b63A new audit runs when the source changes
• Repository statusNot archivedLast push 1 day ago
• Sandboxed runNot runSkills are read, not executed; layer L4 is skipped

Its description: Set up a Vercel API token for publishing apps using browser automation

---
name: vercel-token-setup
description: Set up a Vercel API token for publishing apps using browser automation
compatibility: "Designed for Vellum personal assistants"
metadata:
  icon: assets/icon.svg
  emoji: "▲"
  vellum:
    category: "development"
    display-name: "Vercel Token Setup"
    includes: ["vellum-browser-use"]
---

You are helping your user set up a Vercel API token so they can publish apps to the web.

## Client Check

Determine whether the user has browser automation available (macOS desktop app) or is on a non-interactive channel (Telegram, Slack, etc.).

- **macOS desktop app**: Follow the **Automated Setup** path below.
- **Telegram or other channel** (no browser automation): Follow the **Manual Setup for Channels** path below.

---

# Path A: Manual Setup for Channels (Telegram, Slack, etc.)

When the user is on Telegram or any non-macOS client, walk them through a text-based setup. No browser automation is used - the user follows links and performs each action manually.

### Channel Step 1: Confirm and Explain

Tell the user:

> **Setting up Vercel API Token**
>
> Since I can't automate the browser from here, I'll walk you through each step with direct links. You'll need:
>
> 1. A Vercel account (free tier works)
> 2. About 2 minutes
>
> Ready to start?

If the user declines, acknowledge and stop.

### Channel Step 2: Create the Token

Tell the user:

> **Step 1: Create an API token**
>
> Open this link to go to your Vercel tokens page:
> https://vercel.com/account/tokens
>
> 1. Click **"Create"** (or **"Create Token"**)
> 2. Set the token name to **"Vellum Assistant"**
> 3. Select scope: **"Full Account"**
> 4. Set expiration to the longest option available (or **"No Expiration"** if offered)
> 5. Click **"Create Token"**
>
> A token value will appear - **copy it now**, as it's only shown once.

### Channel Step 3: Store the Token

Tell the user:

> **Step 2: Send me the token**
>
> Please paste the token value into the secure prompt below.

Present the secure prompt (via the bash tool) — it collects the token through a secure UI and stores it with the right tool policy in one step:

```bash
assistant credentials prompt --service vercel --field api_token \
  --label "Vercel API Token" \
  --placeholder "Enter your Vercel API token" \
  --description "Paste the API token you just created on vercel.com" \
  --allowed-tools "publish_page,unpublish_page"
```

### Channel Step 4: Done!

> **Vercel is connected!** You can now publish apps to the web. Try clicking Publish on any app you've built.

---

# Path B: Automated Setup (macOS Desktop App)

You will automate Vercel token creation via the browser while the user watches. The user's only manual action is signing in to Vercel (if needed) and one copy-paste for the token value.

## Browser Interaction Principles

All browser operations are executed through the `assistant browser` CLI, invoked via `host_bash`. Vercel's UI may change over time. Do NOT memorize or depend on specific element IDs, CSS selectors, or DOM structures. Instead:

1. **Screenshot first, act second.** Before every interaction, take a screenshot to see the current visual state. Use `assistant browser snapshot` to find interactive elements.

   ```bash
   assistant browser --session vercel screenshot --output /tmp/vercel-state.jpg
   assistant browser --session vercel --json snapshot
   ```

2. **Adapt to what you see.** If a button's label or position differs from what you expect, use the screenshot to find the correct element.
3. **Verify after every action.** After clicking, typing, or navigating, take a new screenshot to confirm the action succeeded.
4. **Never assume DOM structure.** Use the snapshot to identify what's on the page and interact accordingly.
5. **When stuck, screenshot and describe.** If you cannot find an expected element after 2 attempts, take a screenshot, describe what you see to the user, and ask for guidance.

## Anti-Loop Guardrails

Each step has a **retry budget of 3 attempts**. An attempt is one try at the step's primary action (e.g., clicking a button, filling a form). If a step fails after 3 attempts:

1. **Stop trying.** Do not continue retrying the same approach.
2. **Fall back to manual.** Tell the user what you were trying to do and ask them to complete that step manually in the browser. Give them the direct URL and clear text instructions.
3. **Resume automation** at the next step once the user confirms the manual step is done.

If **two or more steps** require manual fallback, abandon the automated flow entirely and switch to giving the user the remaining steps as clear text instructions with links.

## Things That Do Not Work - Do Not Attempt

These actions are technically impossible in the browser automation environment:

- **Downloading files.** Clicking a Download button via `assistant browser click` does not save files to disk.
- **Reading the token value from a screenshot.** The token IS visible in the creation dialog, but you MUST NOT attempt to read it from a screenshot - it is too easy to misread characters, and the value must be exact. Always use the `assistant credentials prompt` approach to let the user copy-paste it accurately.
- **Clipboard operations.** You cannot copy/paste via browser automation.

## Step 1: Single Upfront Confirmation

Tell the user:

> **Setting up your Vercel API token so we can publish your app...**
>
> Here's what will happen:
>
> 1. **A browser opens** to your Vercel account settings
> 2. **You sign in** (if not already signed in)
> 3. **I create the token** - you just watch
> 4. **One quick copy-paste** - I'll ask you to copy the token value into a secure prompt
>
> Takes about a minute. Ready?

If the user declines, acknowledge and stop. No further confirmations are needed after this point.

## Step 2: Open Vercel and Sign In

**Goal:** The user is signed in and the Vercel tokens page is loaded.

Navigate to `https://vercel.com/account/tokens`:

```bash
assistant brow

Excerpt, first 6,000 characters of SKILL.md, verbatim from the audited commit. It is text from someone else’s repository, quoted so you can judge it before your agent reads it.

git clone https://github.com/vellum-ai/vellum-assistant
git -C vellum-assistant checkout 19bc86343b63

Works in

1. 2026-10-14
   ✓SAFE B 89 / 100

   first audit · source 19bc86343b63

Questions about Vercel Token Setup

• github.com/vellum-ai/vellum-assistant/tree/main/skills/vercel-to — the audited source at commit 19bc86343b63 and every finding aboveObserved · 14 Oct 2026

Spotted something wrong? Report a correction · Methodology