testing-policylisted

# testing-policy — one home for how FFS tests ## Host dispatch contract - Codex: `$skill`, Codex collaboration roles, and GPT-5.6 tiers. - Claude: `/skill`, Agent/Skill tools, and Claude aliases. - A bare `/skill` in this shared source denotes the Claude form; Codex dispatches the same named skill as `$skill`. ## When to invoke - Referenced by `/feature-spec`, `/spec-decompose`, `/feature-implement`, `/code-uplift`, `/fix` when writing or judging tests. Load it before authoring any test plan. - Directly: "what's our testing policy", "how should this be mocked", "what does the browser gate require". ## Why this exists The recurring failure is *green tests, broken browser*: manual testing keeps catching issues automated tests should have caught. Root causes, in observed order: over-mocked modules whose fake shapes drift from reality; jsdom silently dropping layout/CSS/ navigation; no real-browser gate in the agent loop; the agent that wrote the code also grading it. Every rule below exists to close one of those. ## 1. Mock-minimization ladder (stop at the first rung that holds) 1. **Real thing.** Real DB (testcontainers / local Postgres / SQLite where the dialect allows), real filesystem in a tmpdir, real clock unless the test is about time. 2. **Own adapter over anything you don't own.** Third-party API/SDK → wrap in a thin adapter; tests mock YOUR adapter interface; a separate contract test exercises the adapter against the real service (or its sandbox) —