Changelog & Release Notes Scraper

Track what shipped — across your whole stack, in one dataset.

This actor pulls release notes, changelogs, product updates and status incidents from 59 curated sources (AWS, Google Cloud, Azure, Cloudflare, GitHub, Anthropic, Stripe, Shopify, Kubernetes, Linear, Postgres, Chrome…), plus any GitHub repository, package registry or feed you point it at.

Public data only. No account, no API key, no login, and nothing to configure to get started.

Why use this actor

• 59 vendors, one row shape. An AWS bulletin, a GitHub release, an npm version bump and a status incident all come back with the same fields and the same UTC date column — so you can sort, diff and alert across vendors without writing per-vendor glue.
• Not limited to the catalog. Any public GitHub repo (github:redis/redis), any package (npm:react, pypi:django, crates:serde, gem:rails, nuget:…, packagist:…), any RSS/Atom/JSON feed URL — mixed freely in one run.
• Pick a stack in one word. preset: "cloud-providers" or "ai-platforms" or "databases" instead of listing sources by hand.
• Filter where it's cheap. By date (sinceDate), by keyword across title and body, and by count per source.
• Every source reports back. Each one gets a summary row saying how many entries it carried, how many survived your filters, and why — and a source that fails produces an error row rather than vanishing.
• Version-aware. Releases carry a parsed version and an isPrerelease flag that understands semver (1.0.0-rc1), Python/PEP 440 (1.4.0a2) and vendor tags (1.17.0-alpha20260827).

How it works

1. You name the sources you want — catalog keys, a preset, GitHub repos, packages, or plain feed URLs.
2. The actor fetches each one and reads whichever format that vendor publishes.
3. Entries are normalised to one shape, filtered by your date/keyword rules, and written to your dataset newest-first.

You don't manage credentials, scrapers, or browsers.

Input

{
  "preset": "",
  "sources": ["aws", "cloudflare", "github", "anthropic", "github:redis/redis"],
  "maxItemsPerSource": 50,
  "sinceDate": "2026-08-01",
  "keywords": [],
  "includeContent": true,
  "maxConcurrency": 8
}

Field	Type	Description
preset	string	Select a whole family at once: all, cloud-providers, ai-platforms, devtools, databases, languages, browsers-mobile, status-pages, saas, security. Combines with sources.
sources	array	Catalog key (aws), github:owner/repo, npm:react / pypi:django / crates:serde / gem:rails / nuget:Pkg / packagist:vendor/name / hashicorp:vault / eol:nodejs, or any feed URL.
maxItemsPerSource	integer	Newest entries per source. 0 = everything. Default 50.
sinceDate	string	Keep entries on or after this date (YYYY-MM-DD or ISO 8601). Undated entries are kept.
keywords	array	Keep only entries mentioning one of these (case-insensitive) in title, body, version or categories.
includeContent	boolean	Include full body and the untouched upstream object. Turn off for a compact index. Default true.
githubToken	string	Optional. Unlocks full release history for github: sources (see below).
maxConcurrency	integer	Requests in flight at once. Default 8.
minRequestInterval	number	Seconds between request starts. Default 0.

About githubToken

Without a token, github: sources read the repo's public releases feed: no credentials needed, but it serves the 10 most recent releases and cannot page further. With a read-only token they switch to the full API — complete release history, plus draft/pre-release flags and asset lists — and the hourly request allowance rises from 60 to 5,000. Optional, and only affects github: sources.

Output

Two record types share the dataset, told apart by recordType.

ENTRY — one row per release / changelog post / incident

{
  "_input": "cloudflare",
  "_source": "S1-feed",
  "_scrapedAt": "2026-08-31T16:16:17Z",
  "recordType": "ENTRY",
  "sourceKey": "cloudflare",
  "sourceName": "Cloudflare changelog",
  "vendor": "Cloudflare",
  "adapter": "feed",
  "entryRank": 1,
  "title": "Browser Run - Crawl endpoint now respects the Content Signals `use` directive",
  "url": "https://developers.cloudflare.com/changelog/post/2026-08-31-crawl-content-use/",
  "publishedAt": "2026-08-31T00:00:00Z",
  "author": null,
  "summary": "The /crawl endpoint now respects the use directive of the Content Signals standard ...",
  "contentHtml": "<p>The <a href=\"...\"><code>/crawl</code></a> endpoint now respects ... truncated",
  "contentText": "The /crawl endpoint now respects the use directive ... truncated",
  "categories": ["Browser Run"],
  "version": null,
  "isPrerelease": false,
  "raw": { "...": "the upstream item, preserved verbatim" }
}

A GitHub release row from the same run:

{
  "_input": "github:redis/redis",
  "_source": "S1-github-atom",
  "recordType": "ENTRY",
  "sourceName": "redis/redis releases",
  "vendor": "redis",
  "adapter": "github",
  "title": "8.10.1",
  "url": "https://github.com/redis/redis/releases/tag/8.10.1",
  "publishedAt": "2026-08-17T16:46:02Z",
  "author": "YaacovHazan",
  "summary": "Update urgency: SECURITY: There are security fixes in the release. Security fixes (CVE-2026-62356) ...",
  "version": "8.10.1",
  "isPrerelease": false
}

Field	Type	Description
sourceKey / sourceName / vendor	string	Which source this came from.
adapter	string	How it was read: feed, github, registry, markdown.
title	string	Entry headline, or the version/date for release feeds.
url	string	Link to the entry.
publishedAt / updatedAt	string	UTC ISO 8601. Normalised from every upstream date format.
author	string	Where the source provides one.
summary	string	Short plain-text summary; falls back to the opening of the body.
contentHtml / contentText	string	Full body, when includeContent is on.
categories	array	Upstream tags/labels.
version	string	Parsed version. Authoritative for releases and packages; best-effort for prose titles.
isPrerelease	boolean	Alpha/beta/rc/experimental.
entryRank	integer	Position within this source, newest first.
raw	object	The upstream item verbatim, so nothing is lost.

SOURCE_SUMMARY — one row per source

{
  "recordType": "SOURCE_SUMMARY",
  "sourceKey": "cloudflare",
  "feedKind": "rss",
  "feedTitle": "Cloudflare changelogs",
  "entriesAvailable": 1186,
  "entriesReturned": 2,
  "requestsMade": 1,
  "newestEntry": "2026-08-31T00:00:00Z",
  "oldestEntry": "2026-08-30T00:00:00Z",
  "filtersApplied": { "sinceDate": null, "keywords": [], "maxItemsPerSource": 2 },
  "notes": ["truncated_to_2: 1186 entries matched; raise `maxItemsPerSource` for the rest."]
}

notes is the honesty channel — it tells you why a count is what it is (truncated, filtered, or a known upstream limitation) instead of leaving you to guess.

ERROR — one row per source that failed

Every source you ask for produces at least one row, so a source going missing is visible in the data rather than as an absence:

{
  "_input": "totally-bogus-source",
  "_source": "none",
  "recordType": "ERROR",
  "_error": "unknown_source",
  "_errorDetail": "'totally-bogus-source' is not a catalog key, not a `github:owner/repo`, not a `<registry>:<package>` token and not a URL."
}

Known limits

• GitHub without a token gives 10 releases per repo. The public feed offers no pagination. Set githubToken for full history. Flagged in notes.
• NuGet publishes no dates. Its version index is a bare list, so publishedAt is empty for those rows and sinceDate can't filter them. Flagged in notes rather than silently returning nothing.
• version on prose entries is a guess. For releases and packages it comes from the actual tag; for a headline like "Announcing 2.0" it is parsed from text and may be absent or approximate.
• Some feeds are large. Oracle Cloud carries 2,960 entries, Vercel 1,532, npm packages several thousand versions. preset: "all" with maxItemsPerSource: 0 produces a very big dataset — keep the per-source cap on.
• Catalog sources can move. Vendors relocate feeds without redirects; that surfaces as a not_found error row naming the URL, not a silent empty result.