LLM proposes, orbital judges
Llama-3.3-70B authors fresh skills for your task β opinionated SKILL.md bodies with use-it-for, workflow, heuristics, anti-patterns. The orbital engine then classifies each one deterministically.
Llama-3.3-70B writes a fresh set of skills for every task. An open-domain orbital engine derives a physics signature for each one β mass, scope, cross-domain affinity, Lagrange potential β and classifies it into a celestial class (planet, moon, trojan, asteroid, comet, irregular). You get back a deterministically ranked list with a decision rule per skill. Works with Claude Code, Cursor, Windsurf, or any MCP client.
npm install -g meridian-skills-mcp
claude mcp add meridian meridian-mcp
Llama-3.3-70B authors fresh skills for your task β opinionated SKILL.md bodies with use-it-for, workflow, heuristics, anti-patterns. The orbital engine then classifies each one deterministically.
Each generated skill gets a class β planet, moon, trojan, asteroid, comet, or irregular β based on physics features derived from its content (mass, scope, cross-domain affinity).
The MCP client is a 5 KB stdio shim under MIT. Same backend powers the browser miniapp and the npm package. One source of truth.
Llama-3.3-70B (Workers AI) writes 5 candidate skills for your task β full markdown bodies, ~600 chars each, with named tools and decision rules.
For each skill we derive mass, scope, independence, cross_domain, fragmentation, drag, and dep_ratio β from content alone.
Argmax over six per-class scores assigns the celestial class. Final ranking = lexical relevance Γ class boost Γ Lagrange versatility.
Free tier always available. Paid plans lift the per-IP cap and give you a personal monthly budget.
X-Meridian-Calls-Remaining headerBilling handled by Stripe β no card data ever touches our infrastructure. Your API key is shown once on checkout success, stored hashed (SHA-256) on our side. Lost it? Email us for rotation.
Let the router pick the right expert.