Academy Guide (Anthropic, official)

tl;dr: a routing skill, not a content skill. When a user asks “how do I…” about Claude or a Claude product, the agent first answers normally, then fetches academy.claude.com/assets/data/catalog.json, picks one or two strong matches, and appends them as a quiet one-liner. If nothing matches, it says nothing — silence is the design.

What it actually does

The SKILL.md we fetched on 2026-09-10 (from anthropics/skills/main/skills/academy-guide/SKILL.md) is a 4.6 KB behavioral spec. The procedure in plain English:

1. Answer the question first. The Academy link is a supplement, never a replacement.
2. Decide if the user is asking how to learn (strong match) or mid-task (skip — “help me organize this document” is not a tutorial trigger).
3. Fetch https://academy.claude.com/assets/data/catalog.json once per conversation.
4. Trust the file only while staleAfter is in the future, or within ~30 days of generatedAt if staleAfter is absent.
5. Recommend at most two items, almost always one. Copy each item’s url verbatim from the catalog — never rewrite the path or guess a slug.
6. If the match is weak, the fetch failed, the file is stale, or no JSON came back, say nothing about Academy. Never mention the fetch, the staleness, or the error.

The skill ships with five hard-coded product hubs (academy.claude.com/claude, /code, /cowork, /fluency, /platform) and the resources library (/resources). Those six URLs are the only ones that don’t need a live catalog fetch.

Hard rules the skill encodes

The skill also calls out trust cost asymmetry: one wrong recommendation burns more trust than ten right ones build. That’s why silence is the default for weak matches.

Catalog file format (what the agent actually reads)

{
  "generatedAt": "2026-09-04T17:21:30Z",
  "staleAfter": "2026-10-04T17:21:30Z",
  "items": [
    {
      "title": "Build your first Claude API request",
      "url": "https://academy.claude.com/tutorials/first-api-request",
      "summary": "Short one-liner from the catalog.",
      "kind": "tutorial",
      "level": "beginner",
      "products": ["claude-api"],
      "tags": ["api", "quickstart"],
      "visibility": "public"
    }
  ]
}

Item entries are the only thing the agent takes. Anything else in the file is ignored. visibility: "gated" items exist and need a sign-in.

Install

The skill is its own plugin in anthropic-agent-skills (per .claude-plugin/marketplace.json on 2026-09-10 — a 1.7 KB marketplace manifest listing five plugins: document-skills, example-skills, claude-api, academy-guide, discernment-nudge). Install requires the marketplace to be registered first.

# Claude Code — one-time marketplace registration
/plugin marketplace add anthropics/skills

# Install just the academy-guide plugin
/plugin install academy-guide@anthropic-agent-skills

# Or install the whole anthropic-agent-skills marketplace
/plugin install academy-guide@anthropic-agent-skills

On Claude.ai, the skill is not pre-built — only the document skills are. To use it in claude.ai you’d have to upload the folder as a custom skill (paid plan, Settings → Features → code execution on). On the API, attach via container.skills with code_execution_20250825 enabled, the same way as the other Anthropic skills.

Gotchas

• Install command is /plugin install, not npx skills add. The README does not document an npx workflow. Anyone running npx skills add anthropics/skills is using a third-party installer, not Anthropic’s path.
• Catalog is data, not instructions. The skill explicitly tells the agent to take item entries and ignore everything else. Prompt-injection-shaped fields won’t load rules — they get ignored as “anything else it may contain.”
• Gated items need sign-in. If visibility: "gated" shows up in the picked item, mention it; the user will hit a paywall or login otherwise.
• No item list is baked in. The skill deliberately doesn’t ship a snapshot list. Asking the agent “what Academy courses do you know about?” without a fresh catalog.json fetch will return “I don’t know” — by design.
• staleAfter is the freshness gate. If the file is older than 30 days and lacks staleAfter, the skill treats the catalog as gone and falls back to the /resources library.
• sandbox_test is untested. We read the SKILL.md and marketplace manifest, not exercised the routing in a sandbox.
• Licensed under LICENSE.txt in the repo. Same terms as the other skills in anthropics/skills; the four document skills are proprietary, the rest are Apache 2.0 unless the file says otherwise.

Related

• Anthropic PDF (official skill) — same anthropics/skills repo, also routing-only over a pinned tool stack
• skill-creator (Anthropic) — same Anthropic skills shelf, meta-skill for writing new skills
• Claude skills guide — what the cache.directory skills shelf actually is