First-party and practitioner feeds
Labs, product blogs, and technical writers are organized by the section they can inform.
Daily signal for people building agent tools: AI platform moves, research, developer tech, and GitHub changes worth tracking.
How it works
The digest turns a fixed set of trusted sources into a short briefing for people building agent tools. Every stage is explicit, reviewable, and designed to prefer useful signal over volume.
The loop does not search the open web indiscriminately. It starts with source lists kept in the repository, so the material it watches is visible, reviewable, and easy to update deliberately.
Labs, product blogs, and technical writers are organized by the section they can inform.
Open-source and GitHub Enterprise changes are collected as distinct lanes with specific PR, issue, and release links.
Curated accounts and Hacker News help find stories; a canonical source is preferred whenever one exists.
Curated channels can add hands-on agent-building techniques to the appropriate section.
Discovery scripts use only Node builtins. Scripts that invoke an external CLI pass an argument array rather than constructing a shell command.
.github/skills/daily-digest-loop/references/digest-topics.csv.github/skills/daily-digest-loop/references/external-github-repos.json.github/skills/daily-digest-loop/references/internal-github-repos.json.github/skills/daily-digest-loop/references/x-handles.json.github/skills/daily-digest-loop/references/hacker-news.json.github/skills/daily-digest-loop/references/youtube-channels.jsonThe author writes one item per section in the standing order. Multiple related stories can live inside that item, but each story is placed only once.
AI Engineering is a required section. The quality gate rejects unsupported tags and duplicate sections, and requires a From GitHub item to separate open-source and GitHub Enterprise activity. The standing order is part of the writing workflow.
.github/skills/daily-digest-loop/verify-digest-quality.mjs.github/skills/daily-digest-loop/SKILL.md.github/instructions/digest-content-review.instructions.mdCandidate stories are weighed by a simple standard: does this change what someone can build, ship, secure, or afford with an agent? Security issues, tools, integrations, and concrete capability shifts take priority over generic industry noise.
A date, summary, and section items are the source of truth. The publisher updates the browser payload, durable archive, RSS feed, freshness marker, and service-worker cache stamp from that entry.
node .github/skills/daily-digest-loop/publish-digest.mjs --entry day.jsondocs/data/digests.js - the data payload the site loadsdocs/data/YYYY/MM/YYYY-MM-DD.json - a durable daily archivedocs/feed.xml - the RSS and podcast feeddocs/data/version.json - the newest published date for stale-replica recoverydocs/sw.js - the fresh offline-cache identityThe quality check runs before publishing. Link and site checks follow publication and must pass before the change merges. The link check probes ordinary citations while explicitly skipping hosts that cannot be verified this way, such as social, bot-blocked, and GitHub Enterprise links.
A pre-publish check for dates, sections, lanes, citations, and editorial completeness.
A post-publish merge gate that probes citations supported by automated verification.
A post-publish merge gate for data shape, version marker, JavaScript syntax, and local page assets.
The published day remains the common source for the web page, audio edition, Teams card, and InsightsHub industry-news feed. Best-effort delivery failures do not block a correct site publish, but they are surfaced for follow-up.
Today, archive, favorites, and per-day pages all read the published digest data.
The day can be narrated into a daily audio edition and linked from the RSS feed.
A digest card is posted once on the first publish for a date.
An idempotent signed webhook makes the same daily briefing available to its industry-news feed.