GH·RADAR

🎼 Skill Orchestra: evidence-backed skill selection, clear tradeoffs, and delivery checks for agent harnesses.

Claude Code plugin for Unity 6: .meta guards, a compile gate, pre-commit checks and Unity reviewers

MGSV MCP: local documentation research, source browsing and staged converter tools for Metal Gear Solid V. AI-developed with OpenAI Codex under arsenal-pt.

Self-hosted marketplace deal finder: Craigslist, Kijiji, eBay & Facebook Marketplace alerts with AI deal analysis.

Deterministic orchestration engine & visual dashboard for Claude Code CLI. Run multi-agent DAGs, enforce file-scope guards, and use your Pro/Max subscription.

Examples for the Gajo URL shortener API: curl, JavaScript, Python, and MCP setup

A weighted LLM leaderboard for research and engineering work — rank models by what you actually care about, with editable weights

A model-agnostic PPT-generation skill: 5-stage workflow + 15 machine-enforced QA gates + sha256 paper-figure provenance. Kills the 3 classic AI-deck failures (all-text, unverifiable images, opaque process). Ships with a full 8-deck × 4-model benchmark.模型无关的 PPT 生成 Skill：五阶段工作流 + G1–G15 机器门禁 + 论文原图 sha256 溯源，根治 AI 生成 PPT「全是文字、图片来源不可信、流程不透明」三大顽疾。

Measure code by language, folder and purpose; explain classification and compare saved reports across projects and workspaces.

Python/Playwright helper for monitoring the DingTalk Wukong invitation page and assisting activation, with configurable polling, logs, retries, and alerts.

An encrypted secrets vault on your machine that pushes secrets where they need to go (Fly, Vercel, GitHub, .env), with backups after every change.

Independent model benchmarks in your agent: the Artificial Analysis MCP server for Claude, ChatGPT, Cursor and Gemini CLI

Affitor for Claude Code and Gemini CLI: the Affitor MCP server plus the integration skill.

Kachizaka — predictions among friends, points only: scoring rules sealed at kick-off, every point explained, live match facts through Orazaka's MCP connectors, standings in real time.

中文 Agent 文档维护工具包：三个 skills、分层 AGENTS.md、决策记录、Git hooks 与 CI，可按需迁移到其他项目

Free read-only remote MCP server for searching the Socketlist directory of MCP servers

Source-grounded, human-cleared agentic publishing: deterministic provenance gate, cross-lab LLM council, fail-closed publisher.

Python service for audited realtime build notifications

Is this MCP server/tool safe to call right now? Live health checks + alternatives for agents.

Static scale-readiness audit for Node.js/TypeScript repos: finds N+1 queries, unbounded reads, missing indexes, missing timeouts, and other issues that only hurt at scale. CLI + GitHub Action.

List, delete, keep and restore your Claude Code sessions: a trash with restore, a keep list that outlives the 30-day sweep, and a tidy that finds empty and orphaned sessions. Zero tokens.

Fine-tuned Qwen2.5-3B (QLoRA) that turns ML paper abstracts into structured JSON. Scores 0.87 on 150 held-out abstracts vs 0.66 for the untuned base model.

Windows AI Companion — a WPF desktop AI assistant that can see your screen and teach you where to click: chat, screen analysis, visual Guide Me target detection with a teaching overlay, and Whisper voice input. WPF/.NET 8 + Node/TypeScript + Groq.

CI eval & tracing harness for tool-using LLM agents (Claude). Replays a versioned dataset per prompt/model change: deterministic checks on tool calls, actions, cost and latency, then a cached LLM-as-judge. Gates merges in GitHub Actions on pass-rate regressions and records span trees of every LLM/tool call for root-cause tracing.