Weekly Kubernetes Digest

A small Python pipeline that pulls the past week from ~20 curated Kubernetes-and-cloud RSS feeds, asks Claude to synthesize a themed roundup, and writes a markdown post into the blog's content/digest/ directory.

Built on the same pattern as compliance-analyzer — mock-friendly architecture, real LLM, optional API key. Without ANTHROPIC_API_KEY set, the script falls back to a deterministic canned digest from the headlines, so testing and CI don't need credentials.

What it does — and what it deliberately doesn't

Does:

• Pulls RSS feeds only — no HTML scraping, no source bodies, no terms-of-service gymnastics.
• Passes titles + the publisher-provided summary to Claude — not full articles.
• Synthesizes one ~400-word post per week, organized by theme (not by source).
• Hard-codes constraints into the system prompt: never quote >10 words verbatim from any source, every claim must link back, drop pure marketing items.
• Writes a Sources considered this week block at the bottom linking every item that was passed in (cited or not).

Does not:

• Scrape source HTML pages.
• Reproduce article bodies, code blocks, diagrams, or screenshots.
• Run as a "content farm" — the output is a synthesized roundup, with every claim attributed.
• Touch any source whose robots.txt blocks the bot.

The bot identifies itself in fetches with a User-Agent that links to a contact page on the blog: fabiorollin-weekly-digest/1.0 (+https://blog.fabiorollin.com/digest/about/).

Stack

Layer	Choice
Fetch	feedparser
LLM	Anthropic Claude (Sonnet 4.5)
Output	Markdown into a Hugo blog content/digest/
Schedule (Phase 2)	Kubernetes CronJob, weekly
Deploy (Phase 2)	CronJob commits to git → GitHub Actions rebuilds blog

Quick start (local)

pip install -r requirements.txt
cp .env.example .env       # optional — set ANTHROPIC_API_KEY here

# Dry-run to stdout, no file written
python digest.py --dry-run

# Write to ../blog/content/digest/weekly-YYYY-wWW.md
python digest.py

Then rebuild the blog with the existing docker build / push / kubectl set image flow and the new digest will be live at https://blog.fabiorollin.com/digest/.

Phase 2: weekly automation in EKS

The script and Dockerfile are already shaped for a CronJob deployment. The path:

1. Push the blog content folder to a git repo (e.g., github.com/fabiorollin/blog).
2. Set up GitHub Actions on that repo to rebuild + push the blog Docker image and roll the deployment whenever content/digest/*.md changes.
3. Create a deploy key with write access for the digest bot:

   ssh-keygen -t ed25519 -C "digest-bot" -f digest_key -N ""
   # Add digest_key.pub to the blog repo's Deploy keys (write access)

4. Build + push the digest image:

   ECR=807291695385.dkr.ecr.us-east-1.amazonaws.com
   aws ecr create-repository --repository-name weekly-digest --region us-east-1
   docker build -t $ECR/weekly-digest:latest .
   docker push $ECR/weekly-digest:latest

5. Create the secrets in cluster:

   kubectl create namespace weekly-digest
   kubectl create secret generic anthropic-api -n weekly-digest \
     --from-literal=api-key=sk-ant-xxxxx
   ssh-keyscan github.com > known_hosts
   kubectl create secret generic blog-deploy-key -n weekly-digest \
     --from-file=id_ed25519=digest_key \
     --from-file=known_hosts=known_hosts

6. Apply the CronJob:

   kubectl apply -f k8s/cronjob.yaml

The job runs every Monday at 13:00 UTC. The first run can be triggered manually:

kubectl create job --from=cronjob/weekly-digest manual-run -n weekly-digest
kubectl logs -f job/manual-run -n weekly-digest

Cost

• One Anthropic call per week, ~10K input + 600 output tokens with Claude Sonnet ≈ $0.05/week ($2.50/year)
• One CronJob pod, ~30 seconds/week, free inside EKS Auto's existing capacity
• ECR storage: <$1/month
• Net: ~$1–4/month

Source list

Curated in sources.yaml — 20 feeds split across three tiers:

• Tier 1 (5): Kubernetes Blog, CNCF Blog, Last Week in Kubernetes Development, KubeFM, Learnk8s
• Tier 2 (10): Red Hat OpenShift, Datadog, Sysdig, Aqua, Snyk, Grafana, HashiCorp, Buoyant, Solo.io, VMware Tanzu
• Tier 3 (5): Container Solutions, iximiuz, Loft Labs, Ahmet Alp Balkan, KubeWeekly

Edit sources.yaml to add or drop sources. Restart the script — no other config touches needed.

File layout

weekly-digest/
├── digest.py              # main pipeline
├── sources.yaml           # curated RSS feed list
├── requirements.txt
├── Dockerfile             # for the Phase 2 CronJob
├── scripts/
│   └── run.sh             # CronJob entrypoint: clone → run → commit → push
├── k8s/
│   └── cronjob.yaml       # Namespace + CronJob with secret mounts
├── .env.example
├── .gitignore
└── README.md

What this project demonstrates

For a Solutions Engineer / platform engineer / AI-curious portfolio:

• A real LLM content pipeline in production — not a toy, runs on a schedule.
• Cost-bounded prompt design — bounded payload, single round-trip per week, ~$0.05/run.
• Editorial judgment encoded in the system prompt — copyright-aware, signal-over-noise.
• Kubernetes CronJob + git-driven CI/CD — same pattern as a release pipeline, just for content.
• Defensive ops — RSS-only, attribution-required, robots.txt respected, contact path published.