Bump AI SDK packages to v7

[pbakaus]

Summary

Rolls the remaining Dependabot AI SDK major updates into one tested PR instead of landing four separate major-version PRs:

• ai: 6.0.208 in the current lockfile, requested range ^6.0.168 -> ^7.0.14 (7.0.14 resolved)
• @ai-sdk/anthropic: 3.0.85 in the current lockfile, requested range ^3.0.71 -> ^4.0.7 (4.0.7 resolved)
• @ai-sdk/google: 3.0.83 in the current lockfile, requested range ^3.0.75 -> ^4.0.8 (4.0.8 resolved)
• @ai-sdk/openai: 3.0.74 in the current lockfile, requested range ^3.0.53 -> ^4.0.7 (4.0.7 resolved)

Also updates the LLM-backed skill-behavior harness to prefer result.responseMessages, which is the AI SDK v7 continuation-message surface, with a fallback to the previous result.response?.messages shape.

Supersedes Dependabot PRs #321, #322, #323, and #324.

Risk / migration notes

• This is a major AI SDK upgrade family, so provider message/result shapes were the main risk.
• The only code migration needed in-repo was the skill-behavior harness response-message continuation path.
• No generated provider/harness artifacts were refreshed or hand-edited.

Validation

• bun install --frozen-lockfile passed
• bun run build passed
• bun run test passed
• bun run test:skill-behavior passed: 27/27 scenarios across claude-sonnet-4-6, gpt-5.5, and gemini-3.1-flash-lite

I also ran the skill-behavior suite on current main with the old AI SDK stack (ai@6.0.208, providers at v3) as a baseline; it passed 27/27 before this migration.

---

Note

Medium Risk
Major AI SDK versions can change result/message shapes; risk is limited to dev-only LLM harness usage and was mitigated by full build and skill-behavior test runs.

Overview
Bundles a major upgrade of the Vercel AI SDK stack: core ai moves to v7, and @ai-sdk/anthropic, @ai-sdk/google, and @ai-sdk/openai move to v4, with matching lockfile updates (including new transitive deps such as @workflow/serde and removal of the old OpenTelemetry coupling on ai).

The only application code change is in the skill-behavior test harness: multi-turn scenarios now chain conversation context from result.responseMessages (AI SDK v7), with a fallback to result.response?.messages for compatibility.

^{Reviewed by Cursor Bugbot for commit f9fa464. Bugbot is set up for automated code reviews on this repo. Configure here.}

[cloudflare-workers-and-pages]

Deploying impeccable with    Cloudflare Pages

Latest commit:	daa8dc5
Status:	✅  Deploy successful!
Preview URL:	https://b87e0142.impeccable-2rv.pages.dev
Branch Preview URL:	https://codex-ai-sdk-v7-rollup.impeccable-2rv.pages.dev

View logs

[greptile-apps]

Greptile Summary

This PR bumps the Vercel AI SDK packages and updates the skill-behavior harness for the new response shape. The main changes are:

• ai moves to the v7 line.
• @ai-sdk/anthropic, @ai-sdk/google, and @ai-sdk/openai move to their v4 lines.
• bun.lock resolves the matching provider, gateway, and provider-utils updates.
• tests/skill-behavior/harness.mjs now prefers result.responseMessages for multi-turn continuation, with a fallback to the prior result.response?.messages shape.

Confidence Score: 5/5

Safe to merge with low implementation risk after the reported validation.

The change is limited to dependency updates plus one scoped harness migration; AI SDK v7 provides the top-level response-message surface used here, and the fallback keeps the previous shape working.

No files require special attention.

T-Rex Logs

What T-Rex did

• The package manifest and bun.lock were upgraded to ai 7.x and @ai-sdk 4.x, and install, build, and tests were run to verify the changes exited successfully.
• Harness continuation logic was updated to use resultMessages ?? result.response?.messages ?? [], and the probe was saved with logs captured to confirm the behavior and that the run ended with exit code 0.

_{Ran code and verified through T-Rex}

Important Files Changed

Filename	Overview
package.json	Updates AI SDK core and provider devDependency ranges to the v7/v4 major lines; Node ESM setup and zod peer range remain compatible.
bun.lock	Resolves the AI SDK/provider major upgrades and transitive lockfile changes consistently with the manifest.
tests/skill-behavior/harness.mjs	Migrates multi-turn continuation capture to prefer AI SDK v7 result.responseMessages with a backward-compatible fallback.

Sequence Diagram

%%{init: {'theme': 'neutral'}}%%
sequenceDiagram
  participant Test as Skill-behavior scenario
  participant Harness as runTurn()
  participant SDK as AI SDK generateText
  participant Model as Provider model

  Test->>Harness: userPrompt + priorMessages
  Harness->>SDK: "generateText({ system, messages, tools, stopWhen })"
  SDK->>Model: execute model/tool steps
  Model-->>SDK: text + response messages
  SDK-->>Harness: result.responseMessages (v7) or result.response.messages
  Harness-->>Test: responseMessages for next turn

Loading

%%{init: {'theme': 'base', 'themeVariables': {"darkMode": true, "background": "#0d1117", "primaryColor": "#21262d", "primaryTextColor": "#e6edf3", "primaryBorderColor": "#8b949e", "lineColor": "#8b949e", "textColor": "#e6edf3", "edgeLabelBackground": "#161b22", "actorBkg": "#21262d", "actorBorder": "#8b949e", "actorTextColor": "#e6edf3", "actorLineColor": "#8b949e", "signalColor": "#8b949e", "signalTextColor": "#e6edf3", "noteBkgColor": "#373320", "noteBorderColor": "#d4a72c", "noteTextColor": "#f0e6c0", "labelBoxBkgColor": "#21262d", "labelBoxBorderColor": "#8b949e", "labelTextColor": "#e6edf3", "loopTextColor": "#e6edf3", "activationBkgColor": "#30363d", "activationBorderColor": "#8b949e"}}}%%
sequenceDiagram
  participant Test as Skill-behavior scenario
  participant Harness as runTurn()
  participant SDK as AI SDK generateText
  participant Model as Provider model

  Test->>Harness: userPrompt + priorMessages
  Harness->>SDK: "generateText({ system, messages, tools, stopWhen })"
  SDK->>Model: execute model/tool steps
  Model-->>SDK: text + response messages
  SDK-->>Harness: result.responseMessages (v7) or result.response.messages
  Harness-->>Test: responseMessages for next turn

Loading

_{Reviews (1): Last reviewed commit: "Bump AI SDK packages to v7" | Re-trigger Greptile}