$ the-wire · showcase
BREAKING: Vercel SDK Ship Double Breaking Changes
By RepoJournal · Filed · About Vercel · Composed from the cited sources · methodology
![github-actions[bot]](https://avatars.githubusercontent.com/in/15368?v=4)
17 people shipped this
github-actions[bot]
@github-actions[bot]
2 cited
listopad
@listopad
1 cited
felixarntz
@felixarntz
1 cited
github-actions
@github-actions
1 cited
ai-sdk-factory
@ai-sdk-factory
1 cited
pranaygp
@pranaygp
1 cited
QingHeSite
@QingHeSite
bgw
@bgw
Sebastian "Sebbie" Silbermann
@eps1lon
next-js-bot
@next-js-bot
VaguelySerious
@VaguelySerious
dummdidumm
@dummdidumm
Mitul Shah
@mitul-s
vercel-gh-bot-4[bot]
@vercel-gh-bot-4[bot]
benpankow
@benpankow
OwenKephart
@OwenKephart
allenzhou101
@allenzhou101
Two TypeScript SDK releases in two days break response shapes across projects, feature flags, and deployments.
Overnight, Vercel shipped not one but two TypeScript SDK releases, each packed with breaking response changes [1] [2]. `@vercel/sdk` 1.28.21 alters `response` on `featureFlags.createFlag()`, `updateFlag()`, `deployments.cancelDeployment()`, and removes `response.image.vhs` from `vcr.getRepositoryImage()` [1]. The 1.28.20 release similarly reworks `response` across `projects.getProject()`, `updateProject()`, `updateMicrofrontends()`, and `uploadProjectAvatar()` [2]. Both versions are now tagged, so before you bump, audit every call site. Meanwhile, the AI SDK fixed a silent truncation bug: OpenAI-compatible chat streams that end without a finish reason now report as errors instead of succeeding [3]. This closes a gap where truncated SSE streams kept generated text but "appeared to finish successfully," masking incomplete output. Expect `generateImage(...).usage` to finally return real token counts for every provider built on `createOpenAICompatible`, as image generation token usage is now mapped [4]. Over in workflow, `run.returnValue` polling switches from a fixed 1s interval to a World long poll, eliminating quantization latency for already-finished runs [5]. And the Cline harness gets a file read/write bug fix you'll want if you've hit that edge [6]. Also, eve 0.41.0 launches a native Linq channel for iMessage and SMS with guided setup [7]. Prioritize the SDK patch and the stream truncation fix this morning.
Action items
- → Upgrade @vercel/sdk to 1.28.21 and fix breaking response shapes in projects, feature flags, and deployments vercel/sdk [immediate]
- → Update to the latest vercel/ai to get SSE truncation errors and image token usage vercel/ai [immediate]
- → Evaluate workflow run.returnValue long poll change for latency-sensitive paths vercel/workflow [plan]
- → Monitor eve 0.41.0 Linq channel rollout for iMessage and SMS vercel/eve [monitor]
References
- [1] typescript - v1.28.21 - 2026-08-21 04:47:57 ↗ vercel/sdk
- [2] typescript - v1.28.20 - 2026-08-20 16:44:08 ↗ vercel/sdk
- [3] fix: report OpenAI-compatible chat streams without a finish reason as errors ↗ vercel/ai
- [4] feat(provider/openai-compatible): report image generation token usage ↗ vercel/ai
- [5] feat(core): resolve run.returnValue via a World long poll instead of a 1s poll ↗ vercel/workflow
- [6] fix(harness-cline): fix bug with reading and writing files ↗ vercel/ai
- [7] Version Packages ↗ vercel/eve
