mastra-ai / skills

That one skill, called mastra, uses progressive disclosure: a main file that teaches core concepts and routes to reference files on demand — setup and CLI installation, finding APIs already sitting in your own node_modules, fetching current docs from mastra.ai when nothing is installed yet, a common-errors file for troubleshooting, and a migration guide for version upgrades. It also understands the mastra api CLI, for inspecting or calling resources on a local, Mastra Cloud, Trace Intelligence, or remote server.

It is built by Mastra's own team — mastra-ai is the framework's GitHub org — for anyone writing agents, workflows, or tools against Mastra whose APIs move faster than a model's training data can track. It needs nothing beyond a local Mastra project or a network request to mastra.ai: no account, token, or paid plan, which puts it ahead of most of what else needs setup before it does anything.

READ THE FULL ANALYSIS＋

The count used to be higher. Four earlier, more narrowly scoped skills — create-mastra, mastra-best-practices, and two overlapping docs-lookup entries that differ only by a typo in the name — no longer resolve. That matches the current skill's own reference-file layout, which covers the same ground (setup, docs lookup, common errors) inside one file instead of four.

It's discoverable two ways. Beyond installing from GitHub with npx skills add mastra-ai/skills, the repository is also served from Mastra's own .well-known/skills/ endpoint, so any agent that supports that discovery standard can find it directly at mastra.ai without going through GitHub at all.