runbooklisted

# Runbook A runbook is what lets an unattended run get the app to a state where it can verify anything. `visual-verify` cannot screenshot a login screen it does not know how to get past. The rule that separates a runbook from a wish list: **every command in it has been executed from a clean state and observed to work**. An unverified runbook is worse than none, because it costs the reader the time to discover it is wrong. ## What to cover **Preconditions.** Runtime versions, package manager, environment variables and where their values come from, required services (mock backend, database, auth stub), VPN or network access, credentials and who issues them. Name each precondition and how to check it is satisfied — a check is what turns "make sure X is running" into something an agent can act on. **First run from clean.** The literal sequence from a fresh clone to a working app. Actually do it: `git clone` into a temp directory and follow your own instructions. Every step you had to improvise is a step the document was missing. **Reaching states.** How to log in, and as which roles. How to seed or reach test data. How to toggle a feature flag. How to point at a different backend environment. This section is the one agentic verification depends on most and the one runbooks most often omit. **Debugging.** Where the logs are, how to raise verbosity, how to attach a debugger, how to inspect the store or query cache, how to intercept requests. **Failure modes.** The five thin