Writing tests with AI agents

pnpm run client:generate writes, next to the client, an eberly/api/ folder for agents:

• INDEX.md: one line per endpoint, grouped by tag. The one generated for the Nest example:

  <!-- AUTO-GENERATED by eberly. Do not edit. Regenerate: pnpm run client:generate -->
  # API index
  Open `<group>.<method>.md` for details. Call as `user.<group>.<method>(…)`.
  
  ## auth
  - auth.register — POST /auth/register — Register a new user
  - auth.login — POST /auth/login — Log in
  - auth.me — GET /auth/me — Current user
  - auth.whoami — GET /auth/whoami — Current user (deprecated)
  
  ## posts
  - posts.list — GET /posts — List published posts
  - posts.create — POST /posts — Create a post
  - posts.get — GET /posts/{postId} — Get a post by id
  - posts.remove — DELETE /posts/{postId} — Delete a post
  - posts.publish — POST /posts/{postId}/publish — Publish a post
  
  ## comments
  - comments.list — GET /posts/{postId}/comments — Comment tree of a post
  - comments.create — POST /posts/{postId}/comments — Comment on a post
  
  ## users
  - users.uploadAvatar — POST /users/me/avatar — Upload an avatar
  
  ## test
  - test.reset — POST /test/reset — Reset all data

• <group>.<method>.md, e.g. posts.create.md: description, a ready-to-paste call, input, responses by status and the types they use, with field descriptions from the spec.

An agent reads the index, then opens only the endpoints of its scenario: a few dozen lines each, instead of the whole generated.ts (700+ lines for a small backend) or the backend source. Less context spent, fewer made-up fields.

Claude Code skills

Two Claude Code skills use this folder:

• /eberly-setup: asks where the spec is, how auth works and how to reset data, then writes the config, stores, hooks and smoke tests until they are green.
• /eberly-write-tests <scenario>: writes and runs a test for a scenario in plain words.

npx eberly create installs both; npx eberly skills installs them into an existing project.