fix(ai-sdk): keep output validation enabled for tool procedures

[dinwwwh]

createToolFactory disabled both input and output validation on the assumption that the AI SDK re-validates against the tool schemas. That is only true for input: the AI SDK treats outputSchema as type metadata and never validates the value execute returns against it (confirmed in vercel/ai#10222, docs corrected in vercel/ai#11016). Procedures run as tools therefore skipped their .output() schemas entirely, including transforms and defaults.

Fixes

• Output validation stays enabled; only input validation remains disabled, since that half is genuinely redundant.
• A handler returning invalid output now rejects with an output validation error instead of passing through silently.
• Streamed asyncIteratorObject outputs validate every yielded event, and a handler returning a non-iterator now errors instead of being yielded once as the final result (the old fallback branch became unreachable and is removed).

Testing

• A new generateText test with a mock model proves the AI SDK returns schema-violating execute output untouched, so it will start failing if a future ai release adds output validation and makes ours redundant.
• Remaining tests cover input validation being skipped, invalid output rejection, per-event stream validation, and non-iterator rejection.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated (UTC)
orpc	Ready	Preview	Aug 5, 2026 7:45am

[pkg-pr-new]

More templates

• @orpc/bun-playground
• @orpc/cloudflare-playground
• @orpc/nest-playground
• @orpc/next-playground

@orpc/ai-sdk

npm i https://pkg.pr.new/@orpc/ai-sdk@1791

@orpc/arktype

npm i https://pkg.pr.new/@orpc/arktype@1791

@orpc/bun

npm i https://pkg.pr.new/@orpc/bun@1791

@orpc/client

npm i https://pkg.pr.new/@orpc/client@1791

@orpc/cloudflare

npm i https://pkg.pr.new/@orpc/cloudflare@1791

@orpc/contract

npm i https://pkg.pr.new/@orpc/contract@1791

@orpc/experimental-effect

npm i https://pkg.pr.new/@orpc/experimental-effect@1791

@orpc/evlog

npm i https://pkg.pr.new/@orpc/evlog@1791

@orpc/hibernation

npm i https://pkg.pr.new/@orpc/hibernation@1791

@orpc/json-schema

npm i https://pkg.pr.new/@orpc/json-schema@1791

@orpc/nest

npm i https://pkg.pr.new/@orpc/nest@1791

@orpc/next

npm i https://pkg.pr.new/@orpc/next@1791

@orpc/openapi

npm i https://pkg.pr.new/@orpc/openapi@1791

@orpc/opentelemetry

npm i https://pkg.pr.new/@orpc/opentelemetry@1791

@orpc/pinia-colada

npm i https://pkg.pr.new/@orpc/pinia-colada@1791

@orpc/pino

npm i https://pkg.pr.new/@orpc/pino@1791

@orpc/publisher

npm i https://pkg.pr.new/@orpc/publisher@1791

@orpc/ratelimit

npm i https://pkg.pr.new/@orpc/ratelimit@1791

@orpc/server

npm i https://pkg.pr.new/@orpc/server@1791

@orpc/shared

npm i https://pkg.pr.new/@orpc/shared@1791

@orpc/swr

npm i https://pkg.pr.new/@orpc/swr@1791

@orpc/tanstack-query

npm i https://pkg.pr.new/@orpc/tanstack-query@1791

@orpc/trpc

npm i https://pkg.pr.new/@orpc/trpc@1791

@orpc/valibot

npm i https://pkg.pr.new/@orpc/valibot@1791

@orpc/zod

npm i https://pkg.pr.new/@orpc/zod@1791

commit: a5b928a

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[codspeed-hq]

Merging this PR will not alter performance

✅ 25 untouched benchmarks

---

_{Comparing dinwwwh:fix/ai-sdk-output-validation (a5b928a) with main (9f9d664)}

[pullfrog]

✅ No new issues found.

Reviewed changes: Reviewed the full diff of commit a5b928a (the repository's only commit for this PR) in packages/ai-sdk/src/tool.ts and packages/ai-sdk/src/tool.test.ts.

• Output validation re-enabled — createToolFactory now disables only disableInputValidation; disableOutputValidation stays on, so tool procedures validate .output() schemas (including transforms/defaults) that the AI SDK never validates.
• Iterator execute simplified — replaced the isAsyncIteratorObject fallback with yield* await call(...), relying on the (now enabled) asyncIteratorObject validation to reject non-iterators and validate each yielded event.
• New premise-guard test — a generateText + MockLanguageModelV4 test proves the AI SDK passes schema-violating execute output through untouched, and would fail if a future ai release adds output validation.
• Reworked tests — cover input validation being skipped, invalid output rejection, per-event stream validation, and non-iterator rejection.

I traced the behavior through procedure-client.ts (validateOutput/validateInput) and the asyncIteratorObject schema in packages/contract/src/schema-built-in.ts: validate() returns { issues } for non-iterators (→ "Output validation failed") or a wrapped iterator that validates each event (→ "AsyncIteratorObject validation failed"). The new tests map onto exactly these code paths and each genuinely fails under the prior validation-disabled behavior. All 20 tests in tool.test.ts pass and lint is clean. The single caveat—the generateText test uses current AI SDK v4 shapes and is version-fragile test infra—is acknowledged and not a code concern.

This is a correct and well-tested fix that closes a real validation gap.

  ^{｜ View workflow run ｜ Using openrouter/~deepseek/deepseek-v4-flash-latest (free via Pullfrog for OSS) ｜ 𝕏}