[harlan-zw]

🔗 Linked issue

❓ Type of change

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

📚 Description

Tag webhooks crash with Cloudflare error 1101 when the opt-in lookup calls fetch as an object method. Call the injected function directly so opted-in repositories can reach the job queue.

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

[harlan-zw]

🤖 Harlan Agent Kit Agent assessment of head 954190e.

Independent review found no evidence-backed defects. The production exception occurs in readSkillgenOptIns: Cloudflare rejects input.fetch(...) with Illegal invocation. The change calls the injected function directly.

The regression failed before the fix and passes after it. Worker tests, typecheck, lint, and dry-run build pass locally. Production recovery remains unverified.

The Service had three queued Review or Repair tasks and two free host slots. Its Review for this exact head was stopped before the independent assessment. Review confidence: 95/100, with five points deducted for runtime uncertainty.

[harlan-zw]

🤖 Agent production verification for this fix.

The deployed Worker accepted the nuxt-skew-protection tag v1.6.2 replay with HTTP 202. Its existing job had failed on NPM_PROVENANCE_404. After source verification passed, retry job 89163e71-beb7-46c5-82ac-6ede525e069f generated and reviewed the update, then published draft PR #79.

The completed Harness result is Ok: 57 model calls, 545,698 ms, and zero repair attempts. The draft belongs to app/skilld-skillgen, names the exact tag commit, and changes only skills/nuxt-skew-protection/SKILL.md. Maintainer review of the generated examples remains separate.