curating-azure-deployment-learningslisted

# Curating Azure Deployment Learnings The feedback loop that keeps the [diagnosing-azure-deployment-failures](../diagnosing-azure-deployment-failures/SKILL.md) gotcha catalogue current. Capture lessons from real projects in `learnings/`, then promote recurring ones into the catalogue. ## When to invoke - A deployment problem was solved — capture it before it's forgotten - Reviewing accumulated `learnings/*.md` for promotion candidates - Triaging which learnings are project-specific vs broadly reusable ## The pipeline ``` Project hits a problem in the field ↓ Solve it — file a learning in learnings/{date}_{project}.md ↓ Periodically review learnings/ — what shows up across multiple projects? ↓ Promote recurring ones to skills/diagnosing-azure-deployment-failures/references/gotchas.md ↓ Mark the learning entry "✅ Promoted in commit {sha}" ``` ## Learning file format ```markdown --- project: trg-directory-content-crawl date: 2026-05-20 issues: - docker-hub-rate-limit - crawl4ai-pinning severity: medium promoted: false --- # Session Learnings — TRG Directory Content Crawl ## Issue 1 — Docker Hub anonymous pulls hit rate limit in CI ### What happened After 20+ successful deploys, the GitHub Actions runner started getting `toomanyrequests` errors when pulling `unclecode/crawl4ai:latest`. ### Root cause Docker Hub limits anonymous pulls to 100/6h per IP. GitHub-hosted runners share IPs, so the project's pulls were rate-limited from other workflows on the s