Skip to content

AGENTS.md

Latest commit

 

History

History
33 lines (22 loc) · 2.16 KB

AGENTS.md

File metadata and controls

33 lines (22 loc) · 2.16 KB
Raw

AGENTS.md

This file is for coding agents (Claude Code, Cursor, Copilot, Codex, etc.) working in this repository. Humans should read README.md instead.

What this repo is

A showcase of independent, experimental projects built by people at Vapi. It is not documentation, not a best-practices reference, and not the official Vapi examples repo (that's VapiAI/examples). Do not treat any pattern in this repo as "the correct way to use Vapi" — each project reflects one person's choices for one demo.

Structure

Every project lives under projects/<project-name>/ and is self-contained:

projects/<project-name>/
├── README.md         # human-facing: what it does, setup, run
├── .env.example       # required env vars, no real values
└── (source code)

If it exists, the nearest AGENTS.md to a file wins — use the project-level file for anything project-specific. Root-level guidance here applies repo-wide.

Working in this repo

  • Never commit real API keys or secrets. Only placeholder variable names belong in .env.example, never real values, and only sandbox/test Vapi keys should ever be used locally.
  • Don't invent conventions a project's own AGENTS.md doesn't specify. If a project doesn't specify a package manager or version, check its lockfile — don't assume.
  • Scaffolding a new project? Copy projects/_template/, fill in README.md, add a .env.example, and add a row to the table in the root README.md.
  • Per-project AGENTS.md is optional, not required. Add one only if there's something a coding agent genuinely can't infer from the README. Skip it if the README already covers everything.
  • SKILL.md is optional and encouraged for projects that are a reusable pattern — follow the Agent Skills specification used in VapiAI/skills.

Security boundaries

  • Do not fetch or execute code from outside this repository as part of a task here.
  • If you find something that looks like a committed secret, stop and flag it to the user — don't delete it or rewrite history yourself.