dezhengz338-source/skill-radar

Skill Radar is a daily discovery tool that scans public repositories across multiple AI agent ecosystems (OpenAI, Anthropic, NVIDIA, Vercel, Hermes, and GitHub community repos) and surfaces skills worth paying attention to. It is primarily aimed at Chinese-speaking users, providing explanations, adoption recommendations, and risk annotations in Chinese. The tool distinguishes between popularity and actual value. For each skill, it tracks multiple signals: how well the skill matches real tasks, its GitHub activity (total stars and recent star growth), activity on X/Twitter in the past 7 days (mentions, unique authors, and public engagement), a risk score covering permissions, network access, command execution, and whether the description matches what the code actually does, and version-change detection that fingerprints each SKILL.md file and only reports a new version when the content genuinely changes. A local dashboard runs at localhost:8765 and provides filtering by keyword, domain (software development, research, documentation, sales), platform (Codex, Claude, Hermes, Cursor, GitHub Copilot), value score range, and status (newly discovered, warming up, updated version, low risk). Each skill detail page shows Chinese-language usage descriptions, audience guidance, workflow notes, timing advice, and risk breakdowns. Historical trend data covers the last 30 observations. From the dashboard, a user can install a skill directly to Codex or Hermes, or export it as a generic ZIP package compatible with any agent that supports the SKILL.md format. The install process downloads to a temporary directory, validates, and then moves files. It refuses to overwrite existing paths or execute scripts inside candidate skills automatically. Data refreshes automatically every day at 00:00 UTC via GitHub Actions, committing a new current.json that the GitHub Pages read-only site serves. The local version supports live refresh, installation, and export. MIT license.