docs: v2.2 pass — remove huggingface engine refs, update roadmap, sync Chinese docs

- Roadmap: mark P0 (airframe wiring) and P1 (llama/HF strip) as DONE
- Roadmap: add SafeTensors native inference as P2 item
- Roadmap: fix Build Matrix table to reflect current default=[airframe]
- README: fix Linux ARM64 download description, cargo install description,
  build-from-source note (Airframe is public on crates.io), env var table
- docs/zh-CN/README.md, docs/zh-TW/README.md: update feature flag references
- docs/USER_MANUAL.zh-CN.md, docs/USER_MANUAL.zh-TW.md: fix build commands
- shimmy-wiki-content/: update Installation, Quick-Start, Airframe-Engine pages EN+CN+TW

Author: Michael-A-Kuykendall

README.md

@@ -182,6 +182,8 @@ All types are implemented in both the GPU inference shader and a CPU reference i
 
 **Auto-discovery is enabled.** If Shimmy finds GGUF models in your HuggingFace cache, Ollama directory, LM Studio cache (`~/.cache/lm-studio/models`), or local `./models/` folder, it will register and serve them automatically. See [docs/MODEL_EXPANSION.md](docs/MODEL_EXPANSION.md) for the full compatibility matrix.
 
+**SafeTensors format** (`.safetensors`) is supported for model loading — Shimmy routes `.safetensors` files through the `safetensors_native` module. Full inference via the Airframe engine for SafeTensors is a P2 roadmap item (see [docs/v2-roadmap.md](docs/v2-roadmap.md)).
+
 ## 📦 Migrating from v1.x
 
 The llama.cpp backend is **removed in v2.0.0**. The Airframe engine is the only inference path.
@@ -308,7 +310,7 @@ curl -L https://github.com/Michael-A-Kuykendall/shimmy/releases/latest/download/
 # macOS Intel
 curl -L https://github.com/Michael-A-Kuykendall/shimmy/releases/latest/download/shimmy-macos-intel -o shimmy && chmod +x shimmy
 
-# Linux ARM64 (huggingface engine; Airframe cross-compilation not yet supported)
+# Linux ARM64 (Airframe engine; cross-compilation available)
 curl -L https://github.com/Michael-A-Kuykendall/shimmy/releases/latest/download/shimmy-linux-aarch64 -o shimmy && chmod +x shimmy
 ```
 
@@ -317,16 +319,16 @@ curl -L https://github.com/Michael-A-Kuykendall/shimmy/releases/latest/download/
 #### **🛠️ Build from Source / cargo install**
 
 ```bash
-# Install from crates.io
+# Install from crates.io (Airframe GPU engine)
 cargo install shimmy
 
-# Build from source (huggingface engine, no GPU)
+# Build from source
 git clone https://github.com/Michael-A-Kuykendall/shimmy
 cd shimmy
-cargo build --release
+cargo build --release --features airframe
 ```
 
-> **Note**: The Airframe GPU engine is a private dependency and **cannot be built from source** by public users. The [pre-built release binaries](#-download-pre-built-binaries-recommended) already include Airframe compiled in — download those to get full GPU acceleration. `cargo install shimmy` installs the huggingface engine variant from crates.io.
+`cargo install shimmy` or `cargo build --release --features airframe` both give you the full Airframe GPU engine. Airframe is publicly available on [crates.io](https://crates.io/crates/airframe).
 
 ### GPU Acceleration
 
@@ -342,7 +344,7 @@ Release binaries include the Airframe engine with WebGPU support compiled in:
 | **Linux x86_64** | [shimmy-linux-x86_64](https://github.com/Michael-A-Kuykendall/shimmy/releases/latest/download/shimmy-linux-x86_64) | WebGPU (wgpu) | NVIDIA, AMD, Intel |
 | **macOS ARM64** | [shimmy-macos-arm64](https://github.com/Michael-A-Kuykendall/shimmy/releases/latest/download/shimmy-macos-arm64) | Metal (via wgpu) | Apple Silicon |
 | **macOS Intel** | [shimmy-macos-intel](https://github.com/Michael-A-Kuykendall/shimmy/releases/latest/download/shimmy-macos-intel) | Metal (via wgpu) | Intel Mac |
-| **Linux ARM64** | [shimmy-linux-aarch64](https://github.com/Michael-A-Kuykendall/shimmy/releases/latest/download/shimmy-linux-aarch64) | huggingface only | ARM cross-build |
+| **Linux ARM64** | [shimmy-linux-aarch64](https://github.com/Michael-A-Kuykendall/shimmy/releases/latest/download/shimmy-linux-aarch64) | Airframe (WebGPU) | ARM cross-build |
 
 #### **🎯 How GPU Selection Works**
 
@@ -429,7 +431,7 @@ Point your development tools to the displayed port — VSCode Copilot, Cursor, C
 ## 📦 Download & Install
 
 ### Package Managers
-- **Rust**: [`cargo install shimmy`](https://crates.io/crates/shimmy) *(installs huggingface engine; for Airframe GPU, use GitHub Releases binaries)*
+- **Rust**: [`cargo install shimmy`](https://crates.io/crates/shimmy) *(installs Airframe GPU engine)*
 - **VS Code**: [Shimmy Extension](https://marketplace.visualstudio.com/items?itemName=targetedwebresults.shimmy-vscode)
 - **npm**: `npm install -g shimmy-js` *(planned)*
 - **Python**: `pip install shimmy` *(planned)*
@@ -443,10 +445,10 @@ Point your development tools to the displayed port — VSCode Copilot, Cursor, C
 **Full compatibility confirmed!** Shimmy works on macOS with Metal GPU acceleration via wgpu.
 
 ```bash
-# Install from crates.io (huggingface engine)
+# Install from crates.io (Airframe GPU engine)
 cargo install shimmy
 
-# For Airframe GPU engine, download the macOS binary from GitHub Releases:
+# For Airframe GPU engine, you can also download the macOS binary from GitHub Releases:
 curl -L https://github.com/Michael-A-Kuykendall/shimmy/releases/latest/download/shimmy-macos-arm64 -o shimmy && chmod +x shimmy
 ```
 
@@ -508,7 +510,7 @@ I built Shimmy to retain privacy-first control on my AI development and keep thi
 | `SHIMMY_BIND_ADDRESS` | `0.0.0.0:8080` | Full bind address (overrides port) |
 | `SHIMMY_MAX_CTX` | *(from GGUF)* | Override context window; activates YaRN RoPE scaling when above model native |
 | `SHIMMY_MODEL_PATHS` | *(see Zero Config)* | Colon-separated extra model search paths |
-| `SHIMMY_ENGINE_BACKEND` | `airframe` | `airframe` (default) or `llama` (legacy path) |
+| `SHIMMY_ENGINE_BACKEND` | `airframe` | `airframe` (default) |
 | `SHIMMY_ROPE_SCALE` | *(auto)* | Override computed YaRN scale factor |
 | `SHIMMY_KV_QUANT` | `f32` | KV cache quantization: `f32` (default) or `int4` ([TurboShimmy](#-turboshimmy-int4-kv)) |
 | `SHIMMY_PREFILL_CHUNK` | `64` | Tokens per prefill dispatch. Set to `8` on Windows if you see GPU TDR resets on long prompts |

docs/USER_MANUAL.zh-CN.md

@@ -165,21 +165,21 @@ curl -L https://github.com/Michael-A-Kuykendall/shimmy/releases/latest/download/
 chmod +x shimmy
 ```
 
-> **ARM64 注意**：Linux ARM64 版本使用 HuggingFace 引擎。Airframe 的 ARM64 交叉编译支持正在开发中。
+> **ARM64 注意**：Linux ARM64 版本使用 Airframe 引擎，支持交叉编译。
 
 ### 方式二：通过 cargo 安装
 
 ```bash
-# 从 crates.io 安装（使用 HuggingFace 引擎，无需 GPU 即可运行）
+# 从 crates.io 安装（Airframe GPU 引擎）
 cargo install shimmy
 ```
 
 ### 方式三：从源码构建（含 Airframe 引擎）
 
 ```bash
-git clone https://github.com/Michael-A-Kuykendall/shimmy --recurse-submodules
+git clone https://github.com/Michael-A-Kuykendall/shimmy
 cd shimmy
-cargo build --release --features airframe,huggingface
+cargo build --release --features airframe
 
 # 构建完成后，二进制文件位于：
 ./target/release/shimmy
@@ -1109,21 +1109,21 @@ curl http://127.0.0.1:11435/api/health
 ### 构建步骤
 
 ```bash
-# 克隆仓库（含 Airframe 子模块）
-git clone https://github.com/Michael-A-Kuykendall/shimmy --recurse-submodules
+# 克隆仓库
+git clone https://github.com/Michael-A-Kuykendall/shimmy
 cd shimmy
 
-# 构建（仅 HuggingFace 引擎，快速构建，适合 CI）
+# 构建（含 Airframe GPU 引擎，默认，推荐用于正式使用）
 cargo build --release
 
-# 构建（含 Airframe GPU 引擎，推荐用于正式使用）
-cargo build --release --features airframe,huggingface
+# 构建（CPU-only，无 Airframe GPU）
+cargo build --release --no-default-features
 
 # 运行测试
-cargo test --features huggingface
+cargo test
 
 # 安装到系统
-cargo install --path . --features airframe,huggingface
+cargo install --path .
 ```
 
 ### 交叉编译

docs/USER_MANUAL.zh-TW.md

@@ -165,21 +165,21 @@ curl -L https://github.com/Michael-A-Kuykendall/shimmy/releases/latest/download/
 chmod +x shimmy
 ```
 
-> **ARM64 注意**：Linux ARM64 版本使用 HuggingFace 引擎。Airframe 的 ARM64 交叉編譯支援正在開發中。
+> **ARM64 注意**：Linux ARM64 版本使用 Airframe 引擎，支援交叉編譯。
 
 ### 方式二：透過 cargo 安裝
 
 ```bash
-# 從 crates.io 安裝（使用 HuggingFace 引擎，無需 GPU 即可執行）
+# 從 crates.io 安裝（Airframe GPU 引擎）
 cargo install shimmy
 ```
 
 ### 方式三：從原始碼建置（含 Airframe 引擎）
 
 ```bash
-git clone https://github.com/Michael-A-Kuykendall/shimmy --recurse-submodules
+git clone https://github.com/Michael-A-Kuykendall/shimmy
 cd shimmy
-cargo build --release --features airframe,huggingface
+cargo build --release --features airframe
 
 # 建置完成後，執行檔位於：
 ./target/release/shimmy
@@ -1100,21 +1100,21 @@ curl http://127.0.0.1:11435/api/health
 ### 建置步驟
 
 ```bash
-# 複製儲存庫（含 Airframe 子模組）
-git clone https://github.com/Michael-A-Kuykendall/shimmy --recurse-submodules
+# 複製儲存庫
+git clone https://github.com/Michael-A-Kuykendall/shimmy
 cd shimmy
 
-# 建置（僅 HuggingFace 引擎，快速建置，適合 CI）
+# 建置（含 Airframe GPU 引擎，預設，建議用於正式使用）
 cargo build --release
 
-# 建置（含 Airframe GPU 引擎，建議用於正式使用）
-cargo build --release --features airframe,huggingface
+# 建置（CPU-only，不含 Airframe GPU）
+cargo build --release --no-default-features
 
 # 執行測試
-cargo test --features huggingface
+cargo test
 
 # 安裝至系統
-cargo install --path . --features airframe,huggingface
+cargo install --path .
 ```
 
 ### 交叉編譯

docs/v2-roadmap.md

@@ -1,8 +1,8 @@
 # Shimmy v2.x Roadmap
 
 **Last Updated:** 2026-06-07  
-**Branch:** `feature/local-dev-platform-setup`  
-**Current version:** 1.7.4 (publishing toward 2.0.0)
+**Branch:** `release/v2.2-cleanup`  
+**Current version:** 2.1.0 → 2.2.0
 
 ---
 
@@ -19,43 +19,45 @@ historically parked at `archive/llama-cpp-era-v1.9.0`.
 
 ## Roadmap Items
 
-### 🔴 P0 — In Progress
+### 🟢 P0 — Done
 
 **Wire airframe as default inference engine**  
-*Status:* Stub mode (`Backend: Stub mode`) — server starts, models load, no generation  
-*Work:* Connect `InferenceEngineAdapter` → airframe GGUF loading → token generation  
-*Blocks:* Everything else in this roadmap  
+*Status:* ✅ Complete — `InferenceEngineAdapter` wired to airframe GGUF loading and token generation. Server starts, models load, and generation works.  
 *Points:* 5
 
 ---
 
-### 🟠 P1 — Next
+### 🟢 P1 — Done
 
 **Strip llama.cpp + HuggingFace Python bridge**  
-*Status:* Gated but present, causing compile weight  
-*Work (8 points total):*
-- Remove `shimmy-llama-cpp-2` dep and all `#[cfg(feature="llama")]` blocks (5pt)
-- Remove HuggingFace Python subprocess bridge (3pt)
-- Remove `src/engine/llama.rs`, `universal.rs`, `huggingface.rs`
-- Clean `adapter.rs` down to airframe-only path
-- Clean `main.rs` of MoE config, GPU backend selection, llama diagnostics
-- Update CHANGELOG, README, wiki to reflect v2 engine
-
-*Sequence:*
-1. Cargo.toml — remove deps/features
-2. Delete dead engine files
-3. Clean engine/mod.rs
-4. Thin out adapter.rs
-5. Clean main.rs
-6. cargo check → 0 warnings, 0 errors
-7. Docs update
+*Status:* ✅ Complete — dead engine files removed, `adapter.rs` cleaned to airframe-only path, `main.rs` stripped of MoE config and llama diagnostics. `cargo check` passes 0 warnings, 0 errors.  
+*Work completed (8 points total):*
+- Removed `shimmy-llama-cpp-2` dep and all `#[cfg(feature="llama")]` blocks (5pt)
+- Removed HuggingFace Python subprocess bridge (3pt)
+- Removed `src/engine/llama.rs`, `universal.rs`, `huggingface.rs`
+- Cleaned `adapter.rs` down to airframe-only path
+- Cleaned `main.rs` of MoE config, GPU backend selection, llama diagnostics
+- Updated CHANGELOG, README, wiki to reflect v2 engine
 
 *Note:* Users who need llama.cpp have `archive/llama-cpp-era-v1.9.0` on origin.
 
 ---
 
 ### 🟡 P2 — Soon
 
+**SafeTensors native inference**  
+*Status:* Not started — `safetensors_native.rs` module exists with a stub `generate()` method  
+*Work:* Replace stub `generate()` with real inference via airframe. This is the inference path for `.safetensors` files (HF format, no GGUF conversion required).  
+*Design:*
+- `safetensors_native.rs` stub wired into `InferenceEngineAdapter` routing
+- `generate()` delegates to airframe with SafeTensors tensor layout
+- Adds `.safetensors` to the model discovery file filter
+*Points:* 5
+
+---
+
+### 🟡 P2 — Soon
+
 **HuggingFace Hub model sourcing (pure Rust)**  
 *Status:* Not started  
 *Motivation:* The shim should accept HF model IDs, not just local paths.
@@ -121,11 +123,11 @@ Users on the historical llama.cpp path: check out `archive/llama-cpp-era-v1.9.0`
 
 ---
 
-## Build Matrix (Target State)
+## Build Matrix (Current State)
 
 | Command | What you get |
 |---------|-------------|
-| `cargo build` | shimmy with huggingface feature (safe default) |
+| `cargo build` | shimmy with Airframe GPU engine (default features: `airframe`) |
+| `cargo build --no-default-features` | CPU-only build (no Airframe GPU) |
 | `cargo build --features console` | shimmy + local AI dev console |
-| `cargo build --features llama` | shimmy + llama.cpp (historical, still works) |
-| `cargo build --features full` | everything |
+| `cargo build --features full` | shimmy + Airframe + MLX (Apple Silicon) |