chore(ai): use Claude 5.5 models throughout

[harlan-zw]

❓ Type of change

• 📖 Documentation
• 🐞 Bug fix
• 👌 Enhancement
• ✨ New feature
• 🧹 Chore
• ⚠️ Breaking change

📚 Description

Claude calls should use Sonnet 5.5 or Haiku 5.5 throughout. Use Haiku 5.5 for registry derivation and API/CLI calls, and Sonnet 5.5 for Workers AI digest summaries. Normalize harness sampling and thinking for 5.5, and align batch estimates with Haiku 5.5 pricing.

🤖 AI disclosure: Harlan Agent Kit modified this description. My AI open-source policy.

[harlan-zw]

Checked by Harlan Agent Kit, an automated agent.

The sweep found older active Claude models in skilld.dev and Massive Monster. Nuxt SEO already selects Sonnet 5.5.

Cloudflare documents Sonnet 5.5 at https://developers.cloudflare.com/ai/models/anthropic/claude-sonnet-5.5/ but omits Haiku 5.5 from its model catalog. Workers AI calls therefore use Sonnet 5.5 with between-tools thinking. Direct Haiku calls and Studio continuation use Haiku 5.5.

Cloudflare rejected a live model-catalog lookup with authentication error 10000. Request-boundary checks used fixtures, not live inference.

[harlan-github-agent]

🤖 MERGED

Harlan Agent Kit posted this automated review. It is not Harlan's personal review or approval. AI open source policy. Last updated: 2026-10-10 06:46 UTC.

GitHub merged this pull request.

• Logged (55/100): The new thinking param in the Anthropic batch request is not declared in the BatchRequestItem type, breaking the build's typecheck gate while leaving runtime JSON identical. View code Optional follow-up: Add thinking?: { type: 'disabled' } to the params member of BatchRequestItem in layers/registry/server/tasks/ai-generate-submit.ts.
  • Ask an agent to verify and fix this finding
• Logged (20/100): Unverified Haiku 5.5 batch pricing constants would silently skew recorded spend estimates in ai_batch_costs. View code Optional follow-up: Confirm the constants against the Anthropic Haiku 5.5 batch pricing table and cite the source URL in the comment above them.
  • Ask an agent to verify and fix this finding
• Logged (15/100): The pull request's routing diagram image is unavailable (HTTP 400), leaving the stated model-routing plan unverifiable by reviewers. View code Optional follow-up: Re-attach the diagram or replace it with a text routing table (component → model) in the PR description.
  • Ask an agent to verify and fix this finding

3dd9a81e-272f-485b-a1d3-94fbfc95c4f9

Selected findings run after merge and open separate pull requests.

[harlan-github-agent]

Harlan Agent Kit posted this automated review. AI open source policy.

Automated Review: READY.

Read the Review results