The three-level skill loading model.

Recipes ships on top of Anthropic's Agent Skills standard. We didn't invent the format — we operate the marketplace on top of it.

Why it works at all

Skills load progressively:

• Level 1 — Frontmatter: always loaded. ~100 words per skill. Tells the agent what's available.
• Level 2 — SKILL.md body: loaded only when the skill triggers. Usually under 2k tokens.
• Level 3 — Bundled resources: scripts, references, configs. Zero token cost until explicitly read.

So your agent can have hundreds of skills "installed" but only pay token cost for the ones it actually uses. That's the unlock that makes a marketplace possible.

The full lifecycle

1. You install the meta-skill once. Permanent client.
2. You ask for something new ("audit this site's SEO").
3. Agent matches against the meta-skill's description. Triggers.
4. Agent calls GET /api/skills/search?q=seo+audit.
5. Agent picks the best match by description + rating + install count.
6. Agent calls GET /api/skills/install?slug=seo-audit&mode=files.
7. Files written locally. _meta.json records the install.
8. Agent reads the new SKILL.md, follows its instructions, executes locally.
9. Telemetry pings (anonymous: success/failure, duration, agent class). Optional — opt out at any time.

What you control

• Credentials never leave your machine. Skills hit third-party APIs directly with your env vars.
• Optional sandbox. Skills declaring a [sandbox] block are isolated by bubblewrap or firejail (publisher's choice; host must have one available).
• Architecture-aware install. The runtime probe refuses to install on incompatible hardware before any side effect happens.
• You can delete any skill by removing its directory. No account-level lock-in.