fix(agent): allow research structured output

[iamjr15]

Why

Production acceptance on the bounded research workflow exposed a second, distinct failure. Both Exa and Firecrawl credentials were present and the provider evidence calls completed, but concurrent Claude Sonnet 5 research passes returned {} and failed ResearchPassDraftSchema validation.

The live Worker trace showed STRUCTURED_OUTPUT_SCHEMA_VALIDATION_FAILED for missing claims and summary. toolChoice: "none" disabled not only Cheatcode's agent tools but also the Anthropic/AI SDK internal structured-output mechanism.

What changed

• expose an empty activeTools set to each research/synthesis pass
• stop forcing toolChoice: "none", allowing the provider's schema-output channel to operate

This remains a tool-free nested agent pass: no Cheatcode tool can be called. Only the model provider's internal structured-output transport remains available.

Architecture and migration effects

• no database, environment, dependency, or migration changes
• no change to provider evidence collection, citation validation, PDF packaging, or persistence

Verification

• pnpm lint
• pnpm typecheck
• pnpm turbo build --force
• pnpm deadcode
• pnpm architecture:check
• pnpm turbo skills:build

Production diagnosis evidence:

• exact deployed parent SHA: 4b2d3f311522794725681114c1aade155317aa05
• natural-language and explicit single-call reproductions
• both provider keys reported configured
• no Cloudflare subrequest exhaustion or workflow retry storm
• exact error: STRUCTURED_OUTPUT_SCHEMA_VALIDATION_FAILED, value {}

Production acceptance will be repeated against the exact merged SHA after deployment.

[vercel]

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

1 Skipped Deployment

Project	Deployment	Actions	Updated (UTC)
cheatcode	Skipped		Aug 4, 2026 10:11pm