11ai Vercel Sandbox v2 troubleshooting
Separate facts from theories. Preserve exact error, status, resource or run ID, timestamps, version, environment, and original operation classification.
Version baseline: Target @vercel/sandbox 2.9 within major 2 and the current Sandbox CLI. Use persistent-by-default sandboxes, current Node.js and Python images, custom VCR images, multi-user isolation, and v2 lifecycle semantics.
Evidence collection
node -p "require('@vercel/sandbox/package.json').version" 2>/dev/null
rg -n 'Sandbox\.|@vercel/sandbox|VERCEL_OIDC_TOKEN|timeout|runtime|snapshot' . --glob '*.{ts,tsx,js}' --glob '!node_modules' | head -140
rg -o '^[A-Z][A-Z0-9_]*=' .env.example .env.local 2>/dev/null | sed 's/=.*//'
npm run typecheck --if-present
npm test --if-present
npm run build --if-present
Redact VERCEL_OIDC_TOKEN, access tokens, uploaded private files, command environment values, logs containing data, or preview URLs, file content, workflow payloads, command output containing data, and personal information. Do not repeat an operation until idempotency and cost are known.
Classify the failure
- Scope or auth failure — confirm team, project, environment, issuer, and resource ID.
- Lifecycle failure — inspect state transitions, owner, timeout, and cleanup.
- Persistence or retry failure — inspect checkpoints, idempotency, attempts, and version.
- Isolation or network failure — inspect explicit permissions and denied boundary.
- Platform failure — preserve status, usage, logs, and regional evidence.
Remediation discipline
State confidence and missing evidence. Make one bounded change, request approval for remote state, retries, limits, exposure, cancellation, or deployment, then rerun the original check. Never remove isolation, make retries unbounded, or replay non-idempotent work.
Report
Report boundary, evidence, cause or uncertainty, fix, remote and usage impact, data exposure, cleanup, rollback, and verification. If local context is unhealthy, hand off to 11ai-operator-vercel-sandbox-v2-environment.