Agent Skills
› NeverSight/learn-skills.dev
› vercel
vercel
GitHub通过 Vercel REST API 检查项目、部署状态、构建日志及域名。支持查询失败原因、查看实时日志、管理域名配置及执行重新部署操作，适用于排查部署问题或监控项目运行状态。
Trigger Scenarios
用户提到 Vercel 相关术语
部署失败或正在构建
需要查看构建或运行时日志
检查预览 URL 或项目域名
希望检查或重新部署 Vercel 项目
Install
npx skills add NeverSight/learn-skills.dev --skill vercel -g -y
SKILL.md
Frontmatter
{
"name": "vercel",
"license": "Apache-2.0",
"metadata": {
"author": "acedatacloud",
"version": "1.0"
},
"connections": [
"vercel"
],
"description": "Inspect Vercel projects, deployments, build logs and domains via the Vercel REST API. Use when the user mentions Vercel, a deployment that failed \/ is building, build or runtime logs, a preview URL, project domains, or wants to check \/ redeploy a Vercel project.",
"when_to_use": "Trigger when the user wants to list Vercel projects, list recent\ndeployments, read a deployment's build logs to diagnose a failure,\ncheck domains, or trigger a redeploy. The connector stores a Vercel\naccess token with the granted scope; treat env-var values as secret\n(never echo them) and confirm before any redeploy \/ mutation.\n",
"allowed_tools": [
"Bash"
]
}
Call the Vercel REST API with curl + jq. The user's token is in
$VERCEL_ACCESS_TOKEN; every call needs Authorization: Bearer $VERCEL_ACCESS_TOKEN. Base URL: https://api.vercel.com. If the resource is
team-scoped, append ?teamId=$VERCEL_TEAM_ID (set only when the user connected a
team token).
Errors come back as {"error":{"code","message"}} — show message verbatim.
403 forbidden usually means the token can't see that team/project →
re-connect with the right scope.
Helper for the optional team param:
TEAM=""; [ -n "$VERCEL_TEAM_ID" ] && TEAM="?teamId=$VERCEL_TEAM_ID"
AUTH="Authorization: Bearer $VERCEL_ACCESS_TOKEN"
Projects & deployments
# Projects
curl -sS -H "$AUTH" "https://api.vercel.com/v9/projects$TEAM" \
| jq '.projects[] | {name, framework, latestProduction: .latestDeployments[0].url}'
# Recent deployments (optionally filter by ?projectId=… or &state=ERROR)
curl -sS -H "$AUTH" "https://api.vercel.com/v6/deployments${TEAM:-?}&limit=20" \
| jq '.deployments[] | {uid, name, url, state, readyState, created}'
Diagnose a failed build
# Deployment detail
curl -sS -H "$AUTH" "https://api.vercel.com/v13/deployments/DEPLOYMENT_ID${TEAM:+&teamId=$VERCEL_TEAM_ID}" \
| jq '{name, url, state: .readyState, error: .errorMessage}'
# Build / runtime events (the actual logs)
curl -sS -H "$AUTH" "https://api.vercel.com/v3/deployments/DEPLOYMENT_ID/events${TEAM:+?teamId=$VERCEL_TEAM_ID}" \
| jq -r '.[] | select(.type=="stdout" or .type=="stderr") | .payload.text'
Domains & redeploy
# Project domains
curl -sS -H "$AUTH" "https://api.vercel.com/v9/projects/PROJECT_ID/domains${TEAM:+?teamId=$VERCEL_TEAM_ID}" \
| jq '.domains[] | {name, verified}'
To redeploy, POST to https://api.vercel.com/v13/deployments with a
deploymentId (or git source) body — confirm with the user first, it ships
to production.
Gotchas
- Never print env-var values. Listing project env vars returns metadata;
the
/v1/.../env/{id}decrypt route returns secrets — don't call it unless the user explicitly asks, and even then summarize, don't echo. - Deployment
state/readyState:QUEUED → BUILDING → READY | ERROR | CANCELED. created/readyare epoch ms — divide by 1000 for human time.
Version History
- e0220ca Current 2026-07-05 22:56
