follow-builders: an AI builders digest from public sources
follow-builders is an AI news digest skill for curated X posts, podcasts and company blogs; feeds need no user API key, while Telegram or email delivery does.
Skill details
npx skills add zarazhangrui/follow-builders --skill follow-buildersKeeping up with AI builders can turn into a second job: check X, find the newest podcast episodes, then scan company blogs for posts worth reading. follow-builders is an AI news digest skill maintained in the zarazhangrui/follow-builders repository. It reads centrally maintained feeds and assembles the public material into a digest using the reader’s schedule, language and delivery preference. Its README lists 26 AI builders on X, six podcasts and two company blogs, Anthropic Engineering and Claude Blog. That is the project’s current source list, not a promise to cover every AI announcement.
What reaches the digest
On setup, the skill asks whether the user wants daily or weekly delivery, a time zone, English, Chinese or bilingual output, and a destination. A manual /ai request can also fetch a digest on demand. The source list is curated centrally; users cannot add or remove builders and podcasts in their local configuration. That makes the setup simple for readers who want a ready-made list. It is a poor fit if you need to follow a particular lab, company or regional community that the maintainers have not selected.
The Node script prepare-digest.js fetches X, podcast and blog feed files from GitHub’s raw-content host, then reads the relevant prompts. The X prompt asks the agent to skip mundane or purely promotional posts and keep direct links to worthwhile updates. The podcast prompt calls for a concise account of the strongest ideas, including a quotation from the transcript. The blog prompt asks for a title, source link and concrete findings. The output supports English, Chinese and paragraph-by-paragraph bilingual summaries. The digest prompts ask for a direct link to each tweet, episode and article, but no rule drops an item that lacks one.
There is a real gap between those pieces. The README and prepare-digest.js include official blogs, and the JSON output carries a blogs array plus blog prompts. But the current SKILL.md remix steps tell the agent how to process X posts and podcasts, not the blog array. Treat blog coverage as unfinished until the skill instructions are reconciled. The README uses examples/sample-digest.md to show the output format.
Delivery has separate requirements
Fetching the central feeds needs no X or YouTube API key from the user. It does require network access to GitHub’s raw-content host. The scripts run on Node.js, and the README asks users to install the npm dependencies in scripts/. If delivery stays in the terminal or an OpenClaw channel, no separate delivery API key is needed. Telegram requires a bot token and chat ID; email requires a Resend API key and a recipient address. Those credentials are stored in the local ~/.follow-builders/ directory.
The scheduled path matters. OpenClaw can run the skill in an isolated agent session and deliver through a configured channel. For non-persistent agents, the documented system crontab command pipes prepare-digest.js straight into deliver.js. That skips the language model’s remix step, so a scheduled email or Telegram message can contain the raw JSON feed instead of the readable digest produced by /ai. Test this path before relying on it for unattended delivery. The feed and prompts are also fetched remotely, so an update to the central repository can change future output without changing the user’s local skill files.
As of November 12, 2026, GitHub showed 7,120 stars. The README says MIT, but GitHub’s license field is null and the repository has no root LICENSE file, so the license cannot be verified from the repository contents. I’d use this when the central source list is close enough and I’m prepared to check the links and choose a delivery mode. If I need help drafting and reviewing Instagram posts instead of aggregating public sources, instagram-agent-skill writes drafts for a human to post; Follow Builders collects material and assembles a digest. Skip it as a hands-off briefing service until the blog instructions, license and scheduled non-persistent delivery path are clear. This entry is based on the repository’s README, SKILL.md and prepare-digest.js; the scheduled delivery path was not run here.