test_authoring_pglisted

# test_authoring_pg — Postgres test infra Rules live in `test_authoring` — read it alongside. This skill = the test infrastructure PATTERN for Postgres-backed forks. **Your fork's `tests/conftest.py` is the source of truth** for the throwaway DB's naming, what schema artifact seeds it (a live `schema.sql`, a squash, a migration replay), and the fixture roster — read it before writing a test. Everything below is the typical shape, not a contract; where your conftest differs, the conftest wins. A fork may also ship its own superseding test-authoring skill — if one is granted, prefer it. ## Foundation (typical shape) `tests/conftest.py` creates a throwaway Postgres DB at session start, applies the fork's schema artifact, seeds two tenants (Alice / Bob) + a shared system shell, and drives the real app through `TestClient` with real auth. **Key identities (an example roster — confirm against your conftest):** | Name | Kind | ID | |---|---|---| | `USER_ADMIN` | admin user | 1 | | `USER_A` / Alice | tenant user | 10 | | `USER_B` / Bob | tenant user | 20 | | `SHELL_SHARED` | shared system shell | 100 | | `SHELL_A` / `SHELL_B` | per-tenant shells | 101 / 102 | | `PROJ_A` / `PROJ_B` | per-tenant projects | 500 / 501 | | `KEY_A` / `KEY_B` | shell bearer keys | `"ALICEKEY"` / `"BOBKEY"` | **Throwaway DB setup:** - An admin connection (`psycopg.connect(DATABASE_URL_ADMIN, autocommit=True)`) creates a uniquely-named `<fork>_test_<unique>` database at session start and drops it a