Vertex AI Claude Provider Drops response_format — Structured Output Always Fails

What this error means

Stage 1 failed to produce valid 'concerns' array after 3 attempts — aborting review is a Vertex AI / Claude failure pattern reported for developers trying to fix vertex ai claude provider silently dropping response_format field, causing structured json output failures. Based on the imported evidence, treat this as a tool-specific troubleshooting page rather than a generic API error.

Why this happens

Sashiko issue #171: Vertex AI Claude provider (claude-sonnet-4-6 and claude-opus-4-6 in us-east5) drops response_format field in generate_claude(), causing Stage 1 AI review to always fail with ‘failed to produce valid concerns array after 3 attempts’. Direct rawPredict calls with same credentials work fine — the bug is in the SDK layer not forwarding response_format to Vertex ClaudeRequest. 12 comments.

Common causes

• When using Claude models through Google Vertex AI (not direct Anthropic API), the response_format field is silently dropped from requests. This causes any tool or application that expects structured JSON output (via response_format) to fail with parsing errors. The Vertex API endpoint itself works fine for free-form responses, making this a subtle integration bug that only surfaces with structured output requirements.
• Sashiko issue #171: Vertex AI Claude provider (claude-sonnet-4-6 and claude-opus-4-6 in us-east5) drops response_format field in generate_claude(), causing Stage 1 AI review to always fail with ‘failed to produce valid concerns array after 3 attempts’. Direct rawPredict calls with same credentials work fine — the bug is in the SDK layer not forwarding response_format to Vertex ClaudeRequest. 12 comments.

Quick fixes

1. Confirm the exact error signature matches Stage 1 failed to produce valid 'concerns' array after 3 attempts — aborting review.
2. Check the Vertex AI / Claude account, local tool state, and provider configuration involved in the failing workflow.
3. Compare the failing environment with a known working setup, then change one configuration value at a time.

Platform/tool-specific checks

• Verify the command, editor, extension, or API client that produced the error.
• Compare local settings with CI, deployment, or editor-level settings when the error appears in only one environment.
• Avoid deleting credentials, local model data, or project settings until the failing scope is clear.

Step-by-step troubleshooting

1. Capture the exact error message and the command, editor action, or request that triggered it.
2. Check whether the failure is account/auth, quota/rate, model/provider, local runtime, or deployment configuration.
3. Review the source evidence below and compare it with your environment.
4. Apply one change at a time and rerun the smallest failing action.
5. Keep the working fix documented for the team or deployment environment.

How to prevent it

• Keep provider/tool configuration documented.
• Record non-secret diagnostics such as tool version, provider name, model name, and command path.
• Add a lightweight check before CI or production workflows depend on the tool.

Sources checked

• https://github.com/sashiko-dev/sashiko/issues/171

Evidence note: Sashiko issue #171: Vertex AI Claude provider (claude-sonnet-4-6 and claude-opus-4-6 in us-east5) drops response_format field in generate_claude(), causing Stage 1 AI review to always fail with ‘failed to produce valid concerns array after 3 attempts’. Direct rawPredict calls with same credentials work fine — the bug is in the SDK layer not forwarding response_format to Vertex ClaudeRequest. 12 comments.

Related errors

• Vertex AI Claude response_format not supported
• Claude structured output via Vertex AI fails
• Vertex Claude rawPredict vs Anthropic API differences

FAQ

What should I check first?

Start with the exact Stage 1 failed to produce valid 'concerns' array after 3 attempts — aborting review text and the smallest action that reproduces it.

Can I ignore this error?

No. Treat it as a failed Vertex AI / Claude workflow until the root cause is understood.

Is this guaranteed to have one fix?

No. The imported evidence supports the troubleshooting path above, but tool behavior can vary by account, plan, version, provider, and local configuration.

How do I know the fix worked?

Rerun the same command, editor action, or request. The fix is working when that action completes without Stage 1 failed to produce valid 'concerns' array after 3 attempts — aborting review.