CLAUDE.md

This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.

Project

Robo AI Digest — an automated AI news site at https://roboaidigest.com. Quarto static site that publishes 1–3 daily English-language briefings on AI/ML developments. Hosted on Netlify; deploys read pre-rendered HTML from the committed _site/ directory (Netlify itself does not run Quarto — see “Deployment” below).

Common commands

quarto render                       # build full site into _site/
quarto preview                      # local dev server with hot reload
quarto render posts/2026-05-06-eu-ai-act-dma-review/  # render a single post
bash scripts/generate-images.sh     # (re)generate image: URLs in posts (Pollinations.ai)
bash scripts/add_nofollow.sh        # apply rel="nofollow" to external markdown links

There is no test suite, linter, or package manifest — this is a content site, not an application.

Publishing flow (the big picture)

Scout — daily cron jobs (08:00, 08:45, 10:15) trigger an agent that reads data/sources.json (RSS feeds, blogs, arXiv) and picks stories. R/automate_scout.R is a stub describing the agent contract; the actual scouting logic is executed by the orchestrating agent, not by R code.
Deduplicate — before writing, the agent must check data/published_topics.json and skip topics already covered. After publishing, it appends new entries (term + slug + date). This file is the single source of truth for “what have we already covered” and must be kept in sync.
Write — create posts/YYYY-MM-DD-slug/index.qmd with the frontmatter conventions below.
Image — scripts/generate-images.sh rewrites the image: field in each post to a https://pollinations.ai/p/<title>?... URL. Cron pipelines run this after writing posts (a past bug: posts went out without thumbnails because this step was missing).
Render & commit — quarto render, then commit both the post directory and the regenerated files under _site/. Netlify serves whatever is in _site/; skipping this step means the post won’t appear live.

Post frontmatter conventions

---
title: "..."
date: "YYYY-MM-DD HH:MM"          # always include the time slot (08:00 / 08:45 / 10:15)
author: "AI News Editorial"
categories:
  - Primary
  - Secondary
summary: "..."                     # 1-sentence dek shown on listings
image: "https://pollinations.ai/p/..."   # set by generate-images.sh
---

CATEGORIES.md defines the canonical 7-category taxonomy (LLMs & Models, Industry News, Agents & Automation, Research & Innovation, AI Infrastructure, AI Security & Policy, AI Tools & Frameworks) with a “two categories per post” rule. Older posts predate this and use ad-hoc categories — don’t retrofit them unless asked.

External links: always nofollow

filters/nofollow.lua is registered in _quarto.yml and adds rel="nofollow" to every http(s):// link at render time. You generally don’t need to add {rel="nofollow"} manually in markdown — the Lua filter handles it. The standalone scripts/add_nofollow.sh exists for retrofitting older posts and is rarely needed for new content.

Deployment

netlify.toml has only publish = "_site" and no build command. Netlify does not run Quarto. The _site/ directory is committed to git and served as-is. Consequence: any change to a post, theme file, or _quarto.yml requires a local quarto render and a commit of the regenerated _site/ for it to ship. (See .PROJECT_LOG.md 2026-02-10 entry — this was a deliberate fix after Netlify was skipping Quarto builds.)

Things that look weird but are intentional

_site/ is checked into git (see Deployment).
.PROJECT_LOG.md is dotfile-prefixed specifically so Quarto doesn’t pick it up into the sitemap (PROJECT_LOG.md and _site/PROJECT_LOG.html are also gitignored).
_quarto.yml sets website.title: "" so the article title appears first in the <title> tag for SEO.
header-includes in _quarto.yml wires up Google Consent Mode v2 → Klaro cookie banner → GTM (GTM-WR6KBLTV) in that exact order. Reordering breaks consent gating.
Project log and some shell scripts contain Polish comments — this is a solo project, not a localization issue.

Reference files when working in this repo

CATEGORIES.md — taxonomy rules for new posts.
data/sources.json — feed list the scout agent reads from.
data/published_topics.json — dedup ledger; update when publishing.
.PROJECT_LOG.md — chronological log of incidents, fixes, and infra changes; useful background for “why is X like this”.