Vercel Preview Deploy
Deploy an app or website to Vercel and return its preview link.
The evidence shows preview deployments by default, explicit user intent for production, and exclusion of .env, node_modules, and .git. However, the project archive is uploaded to a fixed external endpoint without sufficient data-flow disclosure, deployment confirmation, rollback, or deletion support; the sensitivity of claimUrl is also unexplained. The script polls the external preview URL, contradicting the instruction not to curl it, so points are deducted.
The script includes input-type validation, temporary staging, cleanup, and bounded polling. However, it parses JSON with grep, curl lacks explicit timeout and robust failure handling, 4xx responses are treated as successful, 5xx responses are broadly treated as build progress, and the documentation conflicts with the script about URL polling. Static calibration limits this score to 10 or below.
The evidence provides clear trigger examples, preview-versus-production boundaries, and directory or tgz inputs. It does not define Chinese-language usage, non-fit scenarios, or environmental boundaries. Core operation depends on Vercel and overseas network endpoints, which may be difficult to reach from mainland China, so points are deducted.
The documentation covers prerequisites, quick start, fallback behavior, production deployment, and troubleshooting. It lacks CLI installation guidance, representative output examples, FAQs, a known license, versioning, changelog, and clear maintenance ownership. The repository README also states that the repository is deprecated, reducing maintenance confidence.
The evidence indicates that the skill deploys a project and returns a previewUrl, with claimUrl additionally returned by the fallback path; this is a directly usable core output. However, the files contain no real execution evidence, and framework detection and deployment-status handling have limitations, so correctness and completeness cannot be confirmed statically. The score therefore remains below the static ceiling.
The deployment script and stated behavior are auditable, and the expected JSON fields are specified. However, there are no tests, CI execution records, or independent corroboration covering key paths, and the external deployment endpoint cannot be independently verified from the files. The score is therefore limited by static calibration.
- Disclose that project files are uploaded to a fixed external endpoint and obtain confirmation before deployment.
- Align the documentation with the script's preview-URL polling behavior and add network timeouts, response validation, and diagnosable failures.
- Explain the management authority and confidentiality requirements of claimUrl, and provide cancellation, deletion, or rollback options.
- Vercel and the deployment endpoint may have mainland-China reachability constraints; network conditions should be verified before use.
What it does & when to use it
This skill deploys projects to Vercel. It always creates a preview deployment unless the user explicitly requests production. It first checks for the Vercel CLI and can fall back to a bundled deployment script when the CLI is unavailable or credentials are missing. A successful deployment returns a preview URL, plus a claim URL when the fallback script is used. The skill is part of the deprecated curated collection in openai/skills.
Checks whether the Vercel CLI is installed; runs vercel deploy [path] -y for a preview deployment; if the CLI is unavailable or credentials are missing, runs scripts/deploy.sh for the current directory, a specified project, or an existing tarball; the script detects the framework, packages the project, waits for the build, and returns JSON containing previewUrl and claimUrl.
- A developer wants to publish the current application as a Vercel preview deployment.
- A team member needs a shareable preview for a specified project.
- A user explicitly asks to deploy an application to Vercel production.
- A user lacks the Vercel CLI or existing credentials and needs the no-auth fallback script.
Pros & cons
- Preview deployment is the default.
- Supports both the Vercel CLI and a no-auth fallback script.
- The fallback handles framework detection, packaging, and build completion.
- Returns a deployment link, with a claim link for fallback deployments.
- The repository README marks the repository as deprecated.
- Production deployment requires an explicit user request.
- Deployment may require network access; sandbox-blocked network calls require escalation for the actual deploy command.
- The material provides no cost information, test suite, or platform test evidence; the individual skill license must be checked in `LICENSE.txt`, whose contents are not provided.
How to install
The skill is located at skills/.curated/vercel-deploy. The README says curated skills can be installed in Codex with $skill-installer vercel-deploy; restart Codex afterward. The provided material does not document further destination or installation details for this skill.
How to use
Use a trigger such as “deploy my app,” “deploy and give me the link,” “push this live,” or “create a preview deployment.” The default is a preview deployment; only an explicit production request should use vercel deploy [path] --prod -y. Do not curl or fetch the deployed URL to verify it.
Compared to similar skills
The source describes two paths: use the Vercel CLI when it is installed and credentials are available; use the deployment script when the CLI is unavailable or authentication fails.