An agent skill is a small Markdown file that teaches a tool-using LLM how to do one thing well. Pheidi publishes skills at a well-known path so any agent — Claude Code, Claude Desktop, or your own tooling — can fetch one and immediately know how to work with a Pheidi plan: which endpoint to call, how the data is layered, and how to answer the questions a runner actually asks.

Available skills

SkillWhat it does
read-pheidi-planFetches a plan from the public Plan API and interprets it correctly — race goal, weekly structure, every workout with pace zones, and the vacation / injury / override / redistribution layers that explain why a workout looks the way it does. Read-only.

Using a skill

  • Claude Code — drop the skill folder under .claude/skills/ (or install it however you manage skills). It's user-invocable: give Claude a plan ID, slug, or URL and ask your question.
  • Any LLM — fetch the SKILL.md above, hand it to the model as instructions, then ask about a plan. The skill resolves a plan ID, slug, or full URL, calls the API, and reads the layers in the right order.

Discovery

Skills are listed in a machine-readable catalog at /.well-known/agent-skills/index.json, and the read-pheidi-plan skill is also linked from the API catalog (RFC 9727) under the https://pheidi.training/rels/agent-skill relation. An agent that starts from the API catalog can find the skill without being told where it lives.

Skills, the Plan API, and MCP

The read-pheidi-plan skill is read-only and needs no sign-in — it reads a plan through the public Plan API by its publicId. When you want an assistant that can also act on your account — log workouts, report an injury, block out a vacation — connect the Pheidi MCP server instead, which is authenticated and read/write.

See also