mola

Imported from yberkayinci/mola (AGENTS.md). Install upstream with npx skills add yberkayinci/mola. Copyright stays with the author.

Agent Instructions

Read these before editing:

1. .github/copilot-instructions.md
2. docs/PRODUCT_SPEC.md
3. docs/DATA_SCHEMA.md
4. docs/FINALIZATION.md

Current repository state

main is the release history. There is one working branch pattern and no integration branch:

main
  ^
  | pull request
  |
feature/<small-feature>

For each change:

1. branch from the latest main;
2. keep the branch to one coherent feature or fix;
3. run ./gradlew test and ./gradlew assembleDebug before pushing;
4. perform the smallest emulator/physical-device check that exercises the changed behavior;
5. open a pull request into main and let CI pass before merging.

Never develop directly on main.

Non-negotiable product rules

• Preserve the user-defined threshold: AI never recommends, judges, or changes it.
• Never diagnose addiction or another medical/mental-health condition.
• Never shame, moralize, label the person, or state unsupported causation.
• Crisis-signalling external text must be handled locally and must not enter the normal Gemini/repair path.
• Numeric report facts are computed locally.
• Keep the deterministic fallback usable.
• Keep user profile/records device-local and preserve one-action data deletion.
• Never commit secrets, a real local.properties, or a keystore.properties.
• The app ships with no provider credential. Gemini is reached only through a key the user enters at runtime (AiSettingsStore); never reintroduce a build-time or bundled key.
• LocalAiGateway is the shipped default engine, not a test double. Keep it complete enough to be the product on its own.

Architecture boundaries

• ui/ — Compose product UI
• overlay/ — real system intervention overlay
• permissions/, usage/, monitor/ — Android core
• ai/ — both engines, prompts, policy compiler, validators, settings, safety
• data/ — local models/repository/demo records
• MolaApp.kt — integration/navigation boundary

Do not casually refactor cross-cutting architecture while implementing a small feature. Prefer a narrow change with explicit ownership.

Validation expectations

For Android-core/overlay/voice changes, test the affected path on a physical phone when possible and report device/Android version. For pure logic, add/update unit tests. For AI changes, rerun the relevant golden scenarios in docs/AI_EVALUATION.md, plus offline fallback when network behavior changed.

User-facing copy is Turkish. Code, commit messages, and technical documentation are primarily English.