This daily update tracks GitHub repositories gaining significant momentum, ranking them by percentage star growth over the last seven days to surface emerging tools rather than established giants. This week’s leaders highlight a pivot toward mobile accessibility and cost-efficiency, led by DSH Go’s 8,187% surge in stars for mobile DeepSeek deployment and Microsoft’s TokenEconomizer, which targets API expense reduction for coding agents. The landscape also shows a rise in defensive security and desktop interoperability, exemplified by BLA’s 1,050% growth in adversarial defense tools and Vercel Labs’ Agent Desktop, which enables agents to navigate OS accessibility trees.
| # | Repository | Stars Gained | Gain % | Overview |
|---|---|---|---|---|
| 1 | ccch1mneyyy/dsh-Go | 9334 | 8187.7% | An Android client for DeepSeek Harness that enables running AI models locally on mobile or remotely controlling desktop AI environments. |
| 2 | microsoft/TokenEconomizer | 4390 | 1137.3% | A plugin for coding agents that reduces operational costs by learning repository-specific facts, pruning unused tools, and intervening during repetitive failure loops. |
| 3 | qybaihe/bla | 2080 | 1050.5% | A defensive security tool designed to protect software artifacts from AI-driven reverse-engineering agents by exhausting their resource budgets and deploying deceptive metadata. |
| 4 | vercel-labs/agent-desktop | 13248 | 724.7% | A tool that enables AI agents to interact with desktop applications on macOS, Windows, and Linux by reading the OS accessibility tree as JSON. |
| 5 | emanuele-em/kanzo | 879 | 671% | A lightweight Kanban board and MCP server that enables seamless task synchronization and collaboration between human developers and AI coding agents. |
DSH Go (ccch1mneyyy/dsh-Go )
DSH Go is a community-developed Android client designed to bring the DeepSeek Harness ecosystem to mobile devices. It allows users to run AI models locally via a built-in Linux environment or connect remotely to desktop or server-based instances via SSH. Key features include seamless conversation synchronization between mobile and web, end-to-end encrypted dialogue sharing, and advanced mobile automation capabilities. This tool is ideal for developers and power users who require continuous, portable access to their AI workflows and specialized plugins.
TokenEconomizer (microsoft/TokenEconomizer )
TokenEconomizer is a cost-control plugin designed for AI coding agents like Claude Code, GitHub Copilot CLI, and Codex. It optimizes token usage by implementing micro-skills to remember repository-specific commands, pruning unused tools from the agent’s context, and using a loop guard to redirect agents stuck in repetitive error cycles. Developers can integrate this into their existing workflows to achieve significant reductions in model API expenses without modifying the underlying agent models. The tool is particularly effective for engineers working repeatedly within the same large-scale codebases.
BLA: Blind LLM Agents (qybaihe/bla )
BLA is a defensive security framework built to protect software builds from AI reverse-engineering agents (REA). It employs three primary strategies: poisoning artifacts with decoy metadata, exhausting agent resource limits through complex nesting or large headers, and inducing hallucinations in automated analyzers. The tool supports a wide range of formats, including JavaScript, WASM, SQLite, and EVM bytecode, ensuring runtime behavior remains unchanged while increasing the computational cost for attackers. Developers can integrate BLA into their workflows via CLI or MCP-compatible coding agents like Cursor and Claude Code to harden their released software.
Agent Desktop (vercel-labs/agent-desktop )
Agent Desktop provides a programmatic interface for AI agents to observe and operate desktop software without relying on screenshots or pixel matching. By converting the OS accessibility tree into structured JSON, it allows agents to identify UI elements and perform actions like clicking, typing, and scrolling via a CLI, C library, or MCP server. This tool is designed for developers building autonomous agents that need to navigate complex desktop environments across macOS, Windows, and Linux. It supports both headless operations and physical input via a headed mode.
Kanzo (emanuele-em/kanzo )
Kanzo is a lightweight Kanban board designed to facilitate collaborative task management between humans and AI agents via the Model Context Protocol (MCP). It allows users to manage tasks through a browser-based interface, a terminal command, or directly through AI agents that can move cards and post comments. Key features include real-time local network sharing, Git-based synchronization across machines, and a minimal footprint of just 4.5 MB. This tool is ideal for developers looking to integrate autonomous coding agents into their existing project workflows.