Notes on skills, agents, MCP servers, trust tiers, and how to build with the registry.
When you ship one skill, one agent, and one MCP server that all relate to the same problem, ship them together. Here is the wizard, and what it does under the hood.
Bring your own OpenAI / Anthropic / OpenRouter key, run /compose synchronously, and never let your key touch our database. Here is the data path.
Sign in, write a manifest, paste a markdown body, click submit. The registry takes care of signing, versioning, and discovery.
You have a goal. Someone has probably already published the skill that does it. Here is how to find theirs and skip writing your own.
A guided tour of every surface, what each one is for, and the shortest path to seeing the value yourself.
Every team writing AI agents is rebuilding the same prompts, tools, and connectors from scratch. Amitte is the npm-shaped layer that lets that work compound.
How the three tiers work, what verification requires, and why agents downstream can refuse to run unsigned content.
Describe a goal, the registry searches existing skills, drafts the missing pieces, and hands you a publish-ready bundle. Here's the flow and what's actually happening underneath.
A two-step recipe — copy the snippet, restart Claude Desktop, you are done. Same pattern works for Cursor and VS Code.
A practical decision guide. The three primitives sound similar but solve different problems; this is how to pick.
A skill is a verifiable, signed playbook an agent runs. Here is why they exist, what makes one good, and how the registry surfaces them.