How to use the Article Enrichment agent skill

What the skill makes the agent do

Agent skills are instruction files that an AI agent loads when a request matches them. This one applies to article pages in gbrain that are a wall of raw extracted text under a Content heading with no synthesis. The agent reads the page, gathers context from your other brain pages, and rewrites the page into a structured form.

The rewritten page contains these sections:

• Executive Summary: two or three sentences giving the one thing worth remembering.
• Why It Matters: ties the article to your own projects and interests, using brain context rather than assumptions.
• Quotable Lines: three to five verbatim quotes you can reference later.
• Key Insights: actual insights from the article, not topic labels.
• Surprising or Counterintuitive: what makes this piece different.
• See Also: standard markdown links to related brain pages.

What you need before you start

The skill depends on gbrain. It uses gbrain's query, get and put_page operations to find pages, read them and save changes, so it will not do useful work in a setup without gbrain.

You also need access to an LLM. The skill uses Sonnet by default and can use Opus for high-value content. Articles usually arrive through the media-ingest skill, which creates the raw article pages. That skill is referenced by this one but is not bundled with it, so you will need it separately to create new raw pages.

• A gbrain knowledge base with query, get and put_page available to the agent.
• Access to an LLM: Sonnet by default, Opus optionally.
• The media-ingest skill, if you want new articles to arrive as raw pages automatically.
• No scripts are run by this skill, and it does not use the network itself.

Install the article-enrichment skill

Download the skill ZIP from this page. The package contains a single file, SKILL.md.

For a personal install, unzip the package into ~/.claude/skills/. To scope it to one project, unzip it into that project's .claude/skills/ folder instead.

In the Claude apps, you can instead upload the ZIP under Settings → Capabilities → Skills.

• Personal use across projects: ~/.claude/skills/
• Single project: the project's .claude/skills/ folder
• Claude apps: Settings → Capabilities → Skills, then upload the ZIP

Trigger the skill with natural requests

You do not need a special command. The skill responds to phrases that match its triggers. Ask in plain language, and name the page or the batch when you can.

Trigger phrases from the skill include the following. The skill also applies when a brain page is described as useless, boring or a dump, and when a brain-quality check fails on quotability or actionability for an article page.

• 'enrich this article'
• 'enrich the article'
• 'enriching the article'
• 'enrich brain pages'
• 'batch enrich'
• 'enrich pass'
• 'make brain pages useful'

How the enrichment pass runs

The skill is markdown instructions, not a standalone command-line tool. The agent carries out the work with gbrain's existing operations. The source describes this pipeline: read the page, scan for a raw Content section with no Executive Summary, query the article's key entities for context, restructure the page with the LLM, write the new sections while keeping the raw text in a collapsed block, then add back-links from people and company pages.

For batch work, the agent first finds candidate pages that still carry the needs_enrichment flag, then reads and processes each one. Pages that already have the flag set to false are skipped, so re-running a pass does not reprocess them.

• Find pages still flagged for enrichment.
• Read each page and check its structure.
• Pull brain context for the main entities.
• Rewrite the page and keep the raw source in a details block.
• Clear the needs_enrichment flag in the frontmatter.
• Add back-links from referenced people and companies.

Check the result against the quality bar

An enriched page should pass a short checklist before you rely on it. Look at the quotes first, since they are the riskiest part of the output. Each quote should match the source text word for word.

Sonnet is the default because it suits bulk work, but the source notes that it sometimes paraphrases. For bulk runs the recommended approach is a Sonnet draft pass, a spot-check of about five pages with the LLM-judge brain-quality eval, and a switch to Opus for any batch where quotes are paraphrased.

• Executive Summary with two or three sentences.
• Quotable Lines with at least three verbatim quotes.
• Key Insights with at least three bullets that say something new.
• Why It Matters that refers to specific brain context.
• See Also with standard markdown links, not wiki-links.
• A details block that preserves the raw source.

Safety notes before you run it on real pages

The skill rewrites and overwrites page content and clears the needs_enrichment flag. Because the original text is kept in a collapsed section, nothing is lost, but the page you read is replaced by the new version. Do not run it on pages that must not change automatically.

The skill is designed to avoid overwriting edits made by someone else. If a save returns a revision_conflict, the agent should re-read the page, apply the enrichment to the new text and save against the new revision. If a page returns page_not_found, the agent should search for the title before reporting the article missing.

Quotes are the main accuracy risk. If you need exact wording for legal or academic work, check every quote against the original before use.

• Pages are overwritten, so keep your own backup if a page matters.
• Verify quotes yourself where accuracy is critical.
• The skill runs no scripts and makes no network calls of its own; the LLM and gbrain are the parts doing the work.