rules/adding-settings.md

Rama main · réplica de lectura del repositorio original.

# Adding a New User Setting

When adding a new toggle/setting to the Settings page:

1. Add the field to `UserSettingsSchema` in `src/lib/schemas.ts`
2. For a setting with a permanent built-in default, add its value to `DEFAULT_SETTINGS` in `src/main/settings.ts`. If renderer code also needs the default, export a narrowly scoped, side-effect-free constant from `src/shared/settings_defaults.ts` and reuse that constant in `DEFAULT_SETTINGS`; importing `src/main/settings.ts` pulls electron/node into the renderer bundle.
3. Add a `SETTING_IDS` entry and search index entry in `src/lib/settingsSearchIndex.ts`
4. Create a switch component (e.g., `src/components/MySwitch.tsx`) - follow `AutoApproveSwitch.tsx` as a template
5. Import and add the switch to the relevant section in `src/pages/settings.tsx`
6. Adding a field to `DEFAULT_SETTINGS` breaks the inline snapshots in `src/main/settings.test.ts`. The snapshot helper sorts keys alphabetically, so place a manually added field in alphabetical order or, after confirming the diff is limited to the new default, regenerate with `npm test -- src/main/settings.test.ts -u`.

If the setting adds a built-in default, update the inline snapshots in
`src/main/settings.test.ts`; otherwise `npm test` will fail with
default settings snapshot mismatches.

When adding any setting, decide whether to report it. Telemetry and uploaded chat sessions use allow lists, so a new setting stays unsent until you add it:

- Report non-sensitive settings by default. A setting is non-sensitive if its value is a boolean, a number, or a fixed set of options.
- Never report a setting's raw value if it can hold a credential, account details, a file path, a URL, free text, or a name the user chose. Report a derived boolean instead, such as `hasCustomAppsFolder`.
- To send a setting through both telemetry and chat sessions, add it to `getSharedSettingsDiagnostics` in `src/lib/settingsDiagnostics.ts` and to `DebugSettingsSchema` in `src/ipc/types/misc.ts`. For chat sessions only, add it to `getSessionDebugSettings` and `DebugSettingsSchema`. Use chat sessions only for settings too niche for analytics, since reporters can read a session before sending it.
- Update the pinned property names in `src/lib/posthogTelemetry.test.ts` and `src/lib/settingsDiagnostics.test.ts`. Those tests fail until the lists match, which makes each addition visible in review.

For settings reported through telemetry:

- A setting in `getSharedSettingsDiagnostics` already reaches `getSettingsPersonTelemetryProperties` in `src/lib/posthogTelemetry.ts`. If it has no value in `DEFAULT_SETTINGS`, or for a telemetry-only setting, add it to that function directly, reading it as `settings.myFlag ?? DEFAULT_MY_FLAG`. Define that fallback in the side-effect-free `src/shared/settings_defaults.ts` module. If the setting also has a `DEFAULT_SETTINGS` entry, reuse the constant there so the reported value matches the real default without importing main-process code or evaluating unrelated defaults in the renderer. A setting that stays off until enabled can read `settings.myFlag ?? false` instead. Several branches add properties to this one object at a time, so it conflicts often on rebase — the resolution is almost always to keep both properties, not to pick a side.
- Person properties are delivered as PostHog `$set` events. Keep `$set` in `shouldBypassNonProTelemetrySampling`; otherwise successful settings updates can leave sampled users' person properties stale.

For settings whose default can be overridden remotely:

- Prefer leaving the raw stored field unset until the user explicitly changes it, then compute the effective value as `stored value ?? remote default ?? built-in fallback`. Do not persist remote-applied defaults into `user-settings.json`.

For experiments that may later become enabled by default:

- Add an optional boolean to `BaseUserSettingsFields`, but leave it out of `DEFAULT_SETTINGS` while it defaults off. `undefined` is already falsey; persisting `false` would override a later default of `true` for existing users. Check the effective flag consistently wherever the experiment changes runtime behavior or generated guidance.

For schema-validated settings:

- Assume `UserSettings` and other parsed schema types have already normalized field types. Prefer idiomatic boolean checks like `settings?.flag && !settings.hidden` over defensive literal comparisons like `settings?.flag === true && settings.hidden !== true`, unless you are intentionally handling raw unvalidated persisted data before schema parsing.
- Do not put strict constraints (`.int().min(1).max(128)`) on a stored field without a fallback. One out-of-range or hand-edited value fails the whole `StoredUserSettingsSchema` parse, and `readSettings` then returns `DEFAULT_SETTINGS` for every setting. Override the field in `StoredUserSettingsSchema` only, with `.catch(undefined)` (Zod 4), so only the invalid field becomes `undefined`; readers must then apply the default with `??`, as `maxConcurrentTestRuns` does in `test_run_queue_service.ts`. Keep the strict field in `BaseUserSettingsFields`: `setUserSettings` validates its input with `UserSettingsSchema.partial()`, and a `.catch` there would silently clear the saved value instead of rejecting an invalid update.
- For default-on features that start background work, wait for settings to load
  before applying the field's default. Otherwise a saved `false` still starts
  work during startup; test loading settings followed by the disabled value.
- When a new default hides or blocks existing choices, audit Playwright fixtures
  as well as Vitest mocks. Explicitly opt out in scenarios that still require
  those choices, and update assertions for any warning copy the feature replaces.