Clean, correct one-liner. Using the stable production URL from secrets.INKEEP_AGENTS_API_URL instead of the deployment-specific needs.deploy-agents-api.outputs.url is the right call — restart-scheduler runs after promote, so the request should target the promoted production domain rather than a potentially ephemeral deployment URL.

  ^{｜ View workflow run ｜ Using Claude Opus ｜ 𝕏}

PR Review Summary

(0) Total Issues | Risk: Low

Analysis

This is a correct and well-reasoned fix. The change replaces a dynamic Vercel deployment URL with a stable, secret-based production URL for the scheduler restart endpoint.

Why this is correct:

1. Timing issue fixed: The restart-scheduler job runs after the promote job completes. At that point, the production URL (stored in INKEEP_AGENTS_API_URL) is already serving the newly-promoted deployment. Using the stable production URL ensures the restart call reaches the correct instance.

2. Reliability improvement: The previous approach used the specific deployment URL output (needs.deploy-agents-api.outputs.url), which is a transient Vercel deployment URL (e.g., https://agents-api-abc123.vercel.app). After promotion, the production URL is the canonical entry point, making it more reliable to call.

3. Security preserved: Authorization continues to use the bearer token from secrets (INKEEP_AGENTS_RUN_API_BYPASS_SECRET), maintaining proper access control.

🧹 While You're Here (1) 🧹

🧹 1) vercel-production.yml:9-15 Missing secret documentation

Issue: The workflow header documents required secrets but omits INKEEP_AGENTS_API_URL and INKEEP_AGENTS_RUN_API_BYPASS_SECRET, both used by the restart-scheduler job.

Why: Future maintainers setting up or troubleshooting this workflow may miss configuring these secrets.

Fix: Consider adding to the header comment in a future PR:

# - INKEEP_AGENTS_API_URL: Stable production API URL for scheduler restart endpoint
# - INKEEP_AGENTS_RUN_API_BYPASS_SECRET: Bearer token for internal deployment endpoints

Refs: vercel-production.yml:9-15

Summary: Clean, correct fix. The change from dynamic deployment URL to stable production URL is appropriate for the scheduler restart endpoint that runs post-promotion. No blocking issues found. 🚀