Skip to main content

​
Agent skills & rules

Two installable guardrails teach your AI coding agent how Flynet works before it writes a line — the agent skill for Claude Code, and rules for Cursor. Both encode the same thing: the two-credential auth model, the API rules that trip up first integrations, and how to compose @flynetdev/core and @flynetdev/react.

​
What they teach

  • The two route families and their credentials — X-API-Key for Discovery, Bearer JWT for member routes — and that they’re never interchangeable.
  • /users/me/* is canonical; there’s no /users/{id}/*.
  • PKCE is required on authorize; scope mismatches return 403, not 404; unknown filters are silently ignored; there are no payment webhooks in v1.
  • Compose the real @flynetdev/react components; keep the API key and client_secret server-side.

​
Install

One command installs both from npm, into the project you run it in: 
npx -y @flynetdev/skills
That writes .claude/skills/flynet/SKILL.md (Claude Code discovers it automatically — works in Claude Code and Claude Desktop) and .cursor/rules/flynet.mdc (Cursor attaches it when you edit .ts/.tsx files). The files are copied into your project, so they travel with your repo; re-run the command any time to update. Want just one? 
npx -y @flynetdev/skills claude   # skill only
npx -y @flynetdev/skills cursor   # rules only

​
Pair them with the MCP servers

The skill and rules front-load the approach; pair them with the MCP servers for the specifics and execution. The Docs MCP grounds an exact field or status code at the moment the agent needs it, and the API MCP lets the agent call Flynet directly.

​
Next

Set up the MCP servers →

How it fits together →