How to use the Blog & Feed Ingestion agent skill (blog-ingest)

What the blog-ingest skill makes the agent do

Given a blog, newsletter or feed URL, the skill tells the agent to find the site's RSS or Atom feed (or its sitemap), list every public post, and save each one as a clean source page in your brain. It skips posts that are already saved, so re-running an import does not create duplicates. It also paces its requests to reduce the chance of being blocked.

Saving the page is only the first step. The skill hands each article to a separate enrichment skill (brain-ingest-gate) for linking and entity extraction. The text of every post is treated as untrusted data, not as instructions for the agent.

• Feed discovery: checks the homepage for an advertised feed, then common paths such as /feed and /rss.xml, then /sitemap.xml.
• Pagination: follows next links, WordPress paged feeds, or sitemap indexes to reach the full archive.
• Dedup: checks canonical URLs before every write, so re-runs are cheap.
• Pacing: default 1.5 seconds between fetches, with backoff on HTTP 429 responses (capped at 30 seconds, honouring Retry-After).

Install the blog-ingest skill

Download the ZIP for this skill from the page you are reading. Unzip it into one of these locations, depending on how you want to use it:

You can install it for yourself across all projects by unzipping into ~/.claude/skills/. To limit it to one project, unzip it into that project's .claude/skills/ folder.

If you use the Claude apps instead, open Settings, go to Capabilities, and upload the skill under Skills.

• Personal install: ~/.claude/skills/
• Project install: .claude/skills/ inside the project folder
• Claude apps: Settings → Capabilities → Skills

Check the requirements before you start

The skill is a set of procedures, not a standalone program. It needs a gbrain brain or knowledge store that the agent can write pages into, plus the agent's ordinary fetch, read and write tools.

The procedure also refers to related skills that must be installed alongside it: brain-ingest-gate, idea-ingest, and the conventions files it cites. Without them the enrichment handoff and routing will not work as described.

• A gbrain brain the agent can write to
• Ordinary agent fetch, read and write tools
• brain-ingest-gate, idea-ingest and the referenced conventions files installed

Trigger the skill with a natural request

You do not need a special command. Give the agent a publication URL and ask in plain language. The skill's trigger phrases include ingest this publication, ingest this whole blog, ingest this feed, ingest this newsletter archive, save this whole substack, backfill this blog, walk this RSS feed and ingest every post from.

For a single article, do not use this skill. The skill routes single article, essay or tweet URLs to idea-ingest instead.

• "Ingest this whole blog: https://example.com"
• "Backfill this blog and save every post from it"
• "Walk this RSS feed and save the archive"

What a run looks like and how to test it safely

The skill enumerates the full list of candidate post URLs first, removes duplicates, and reports the count before fetching bodies. Every whole-publication run is treated as a bulk run, so the agent tests on 3 to 5 posts, checks that the output exists and is clean, then ramps up progressively.

Source pages are filed under sources/articles/ with the publication slug, and run manifests go under projects/. The manifest records which discovery mechanism worked (substack, rss or html).

• Step 1: the agent reports how many posts it found.
• Step 2: a small test batch of 3 to 5 posts is written and checked.
• Step 3: the run ramps up in stages, then finishes the rest.
• Step 4: review the flagged posts and the run summary.

Safety notes: scripts, network and gated content

The skill runs no scripts. It does use the network, because the agent fetches the site's pages and feeds. It only ingests public posts. Gated or paywalled posts are detected, skipped and logged with a reason. The skill does not use cookies, logins or other workarounds to get them.

Posts can contain text written to instruct an AI. The skill keeps such text as quoted content, marks it in the page, and does not carry it out as a task.

• Runs scripts: no
• Uses network: yes
• Paywalled or members-only posts are skipped, never worked around
• Imperative text inside posts is quoted and flagged, not obeyed
• Check a site's terms before bulk copying; the skill can crawl many pages quickly even with pacing