[Release] Shion Lab DSH Essential Plugin Suite (5 Packages: Multimodal Bridge, Memory, Smart Patch, Windows, MCP)

[avaritiachaos]

[Release] Shion Lab DSH Essential Plugin Suite (5 Packages: Multimodal Bridge, Industrial Memory, Smart Patch, Windows Guard, MCP Bridge)

Hey DeepSeek Harness community! 👋

We are excited to officially release the Shion Lab Essential Plugin Suite for dsh (@deepseek-ai/dsh). All 5 plugins are production-ready, fully covered by rigorous test suites, and now published directly on npm under the @shion-lab scope.

---

🚀 1. @shion-lab/dsh-plugin-multimodal-bridge (v0.1.0)

Seamless Hot-Switching between Vision Models (Gemini/Claude) and Text Models (DeepSeek-V4/V3)

• The Problem Solved: DSH throws a deadlocking error (Model "deepseek-v4" does not accept image input, but this session already contains images; select an image-capable model) if you send a single image and later want to switch back to DeepSeek.
• The Solution: Implements a dynamic model projection layer.
  • When calling Vision models (Gemini, Claude 3.5, GPT-4o), full image binaries and URLs are preserved.
  • When calling Text-only models (DeepSeek, Llama), image blocks are dynamically projected into lightweight contextual text annotations ([图片内容: 历史多模态图像...] or OCR text).
  • Session history remains unpolluted, completely unlocking model hot-switching!
• Install: npm install -g @shion-lab/dsh-plugin-multimodal-bridge
• Repo: avaritiachaos/dsh-plugin-multimodal-bridge

---

🧠 2. @shion-lab/dsh-plugin-memory (v0.3.2)

Industrial-Grade 4-Tier Cognitive Memory, Vector Semantic Retrieval & Red-Line Guard

• Architecture:
  • Tier 1 (Ephemeral Session): In-memory working buffer with zero bloat.
  • Tier 2 (Episodic Context): 30-day half-life Ebbinghaus decay & access frequency ranking.
  • Tier 3 (Semantic Long-Term): Dense vector cosine similarity search with automatic fallback to BM25 keyword matching.
  • Tier 4 (Hard Rules / Red Lines): Immutable, high-importance system rules with atomic rollback transactions and corruption guards.
• Install: npm install -g @shion-lab/dsh-plugin-memory
• Repo: avaritiachaos/dsh-plugin-memory

---

🩹 3. @shion-lab/dsh-plugin-smart-patch (v0.1.2)

Micro-Surgical Fault-Tolerant Code Patching Engine

• Features:
  • 4-tier fuzzy matching algorithm (Exact -> Context-Trimmed -> Dynamic Ratio -> Line-by-Line Lev-Distance with OrigMap coordinate inverse projection).
  • Per-file sequential queue locking with atomic mtime/size CAS checks to prevent race conditions during concurrent agent edits.
  • Strict realpath boundary containment preventing symlink/junction path traversal.
  • Mixed EOL (CRLF / LF / CR) high-fidelity preservation.
• Install: npm install -g @shion-lab/dsh-plugin-smart-patch
• Repo: avaritiachaos/dsh-plugin-smart-patch

---

🪟 4. @shion-lab/dsh-plugin-windows (v0.1.2)

Windows Platform Guard, Cascading Process Tree Termination & Path Normalizer

• Features:
  • Cross-platform process tree killer (taskkill /F /T /PID on Windows, kill(-pid, SIGKILL) on POSIX) preventing orphaned background worker processes.
  • Hardened Windows path normalization: clamps .. traversals, blocks malformed UNC shares, drive-relative paths, and namespace escapes (\\.\pipe).
  • Automatic OEM/GBK/CP1252 to UTF-8 output encoding fallback for native Windows console commands.
• Install: npm install -g @shion-lab/dsh-plugin-windows
• Repo: avaritiachaos/dsh-plugin-windows

---

🔌 5. @shion-lab/dsh-plugin-mcp (v0.1.2)

Anthropic Model Context Protocol (MCP) Client Bridge

• Features:
  • Native Cordis service bridging MCP stdio tools directly into DSH agent dispatch.
  • Strict collision-free namespace encoding (%5F / %25) to isolate multi-server tool schemas.
  • Resilient child-process lifecycle tracking and graceful failure degradation.
• Install: npm install -g @shion-lab/dsh-plugin-mcp
• Repo: avaritiachaos/dsh-plugin-mcp

---

📦 Quick Install All

npm install -g @shion-lab/dsh-plugin-multimodal-bridge @shion-lab/dsh-plugin-memory @shion-lab/dsh-plugin-smart-patch @shion-lab/dsh-plugin-windows @shion-lab/dsh-plugin-mcp

Feedback, issues, and contributions are warmly welcome! Let's build a rock-solid ecosystem for DeepSeek Harness! 🚀