Updated PromptFormatter and LlamaChatSessionImpl to support explicit assistant turn lifecycles and thinking-block extraction.

*   **SDK**:
    *   Enhanced `PromptFormat` and `PromptFormatter` to handle assistant prefixes/suffixes separately and support `<think>` tags.
    *   Updated `LlamaChatSessionImpl` to leverage native EOG token decoding for turn closing, reducing Kotlin-side logic.
    *   Removed `systemPromptReserve` from `DecodeConfig` and `LlamaSession` in favor of simplified context management.
    *   Fixed `PromptFormatter` to conditionally add BOS tokens based on the model's chat template.
    *   Reduced default `contextSize` to 2048.
    *   Added Android logging utilities to the native C++ layer.

*   **App**:
    *   Redesigned `ChatScreen` UI with a new message input style, copy-to-clipboard functionality, and improved markdown rendering.
    *   Added support for displaying thinking blocks in the chat history.
    *   Updated `AppNavigation` to reactively handle model loading states.
    *   Modified `ChatViewModel` system prompt and message mapping.

Author: whyisitworking

CLAUDE.md

@@ -0,0 +1,105 @@
+# CLAUDE.md
+
+This file provides guidance to Claude Code (claude.ai/code) when working with code in this repository.
+
+## Project Overview
+
+**Llama Bro** is an Android SDK for on-device LLM inference, wrapping [llama.cpp](https://github.com/ggerganov/llama.cpp) via JNI. It consists of two modules:
+- **`:sdk`** — reusable Android library (published to JitPack)
+- **`:app`** — demo application showcasing the SDK
+
+The llama.cpp engine is vendored as a git submodule at `sdk/src/main/cpp/external/llama.cpp`.
+
+## Build Commands
+
+```bash
+# Initialize submodules (required after clone)
+git submodule update --init --recursive
+
+# Build SDK AAR
+./gradlew :sdk:assembleRelease
+
+# Build demo app (debug)
+./gradlew :app:assembleDebug
+
+# Install debug app on connected device
+./gradlew :app:installDebug
+
+# Run unit tests (SDK only; no instrumentation tests exist)
+./gradlew :sdk:test
+
+# Run a single test class
+./gradlew :sdk:test --tests "com.suhel.llamabro.sdk.util.PromptFormatterTest"
+
+# Publish SDK to local Maven repository (used by JitPack)
+./gradlew :sdk:publishToMavenLocal
+
+# Clean
+./gradlew clean
+```
+
+**NDK requirement:** NDK 29.0.14206865 and CMake 3.22.1 must be installed via the Android SDK Manager. The project only builds for `arm64-v8a` — x86_64 emulators are not supported.
+
+## Architecture
+
+### Layer Stack
+
+```
+UI (Jetpack Compose)
+  ↓
+ViewModels (MVVM, Hilt-injected)
+  ↓
+Repositories (ChatRepository, ModelRepository)
+  ↓
+SDK Public API  ─────────────────────────────────────────────
+  LlamaEngine → LlamaSession → LlamaChatSession
+  ↓
+JNI Bridge (llama_engine_jni.cpp, llama_session_jni.cpp)
+  ↓
+Native C++ (session.cpp, engine.cpp → llama.cpp)
+```
+
+### SDK Module
+
+Three tiers of API, each building on the previous:
+
+1. **`LlamaEngine`** — loads a GGUF model file from disk; creates sessions. Use `LlamaEngine.createFlow(modelConfig)` for reactive loading that emits `ResourceState<LlamaEngine>`.
+
+2. **`LlamaSession`** — low-level token control. Call `setSystemPrompt()`, then loop `prompt()` + `generate()` to produce tokens. Wrap using `createChatSession()` to get the high-level API.
+
+3. **`LlamaChatSession`** — high-level conversational API. `completion(message)` returns `Flow<Completion>`. Handles prompt template formatting, thinking-block extraction (`<think>...</think>`), and `OverflowStrategy`.
+
+**`ResourceState<T>`** is the lifecycle ADT used throughout. It has subtypes `Loading(progress)`, `Success(value)`, `Failure(error)` and rich Flow extension operators (`flatMapResource`, `filterSuccess`, etc.) for composing resource loads.
+
+**`PromptFormat`** / **`PromptFormats`** — chat template definitions. Built-in formats: `Llama3`, `Gemma3`, `ChatML` (Qwen/Yi), `Mistral`. Each model in `ModelZoo` references one of these.
+
+**`LlamaError`** — sealed error hierarchy (`ModelNotFound`, `ModelLoadFailed`, `ContextOverflow`, `DecodeFailed`, `Cancelled`, `NativeException`, etc.).
+
+### App Module
+
+Standard MVVM with Hilt DI:
+
+- **`ModelRepository`** — singleton managing model download/load/eject lifecycle. Download state is a FSM: `NotDownloaded → Downloading → Downloaded`. The currently loaded engine is exposed as `currentInferenceContextFlow: StateFlow<CurrentInferenceContext?>`.
+- **`ChatRepository`** — Room-backed CRUD for conversations and messages.
+- **`ModelZoo`** — hardcoded list of 6 pre-curated GGUF models (SmolLM2 135M–1.7B, Qwen2.5 0.5B, Llama-3.2 1B, DeepSeek-R1 1.5B) with download URLs and default configs.
+- Navigation uses type-safe `Route` sealed class with Jetpack Navigation Compose.
+
+### JNI / Native
+
+- `sdk/src/main/cpp/jni/` — JNI entry points that convert Kotlin calls to C++ and map C++ exceptions to `LlamaError` subtypes via `NativeErrorMapper`.
+- `sdk/src/main/cpp/session.cpp` — C++ session implementation wrapping `llama_context`.
+- OpenCL/GPU is intentionally disabled (causes UI stalls on mobile). OpenMP multi-threading is enabled.
+
+## Key Configuration Classes
+
+| Class | Purpose |
+|---|---|
+| `ModelConfig` | Model path, `PromptFormat`, MMAP/MLOCK flags, thread count |
+| `SessionConfig` | Context size, `OverflowStrategy`, `InferenceConfig`, `DecodeConfig` |
+| `InferenceConfig` | Temperature, top-p/k, min-p, repeat penalty |
+| `DecodeConfig` | Batch sizes for performance tuning |
+| `PromptFormat` | Per-role prefix/suffix tokens, BOS/EOS, `<think>` tag markers |
+
+## Testing
+
+Unit tests live in `sdk/src/test/` — currently only `PromptFormatterTest` covering chat template formatting. There are no instrumentation tests. New SDK behavior should be covered in this test source set.