Audit AI skill safety from your browser · AI Skill Safety Scoreboard

Paste any SKILL.md, Codex skill.yaml, Claude Code command, or OpenCode config — drop a URL, or pick a sample. The full static / semantic / adversarial pipeline runs in your browser session, free.

0 chars

Free, no key, no sign-in. The static layer catches most of it.

No key needed · Google AI quota provided by the host · limited free runs

Use my own API key (unlocks the semantic + adversarial layers)

Include semantic LLM pass Include adversarial prompt fuzz

Without a key we run static rules only. Findings still cover ~70 percent of common issues.

Paste a SKILL.md and click Run audit to see findings here.

Use it inside Claude Code via MCP

TAR Engine ships an MCP server. Add it to Claude Code in three commands and call `/audit-skill` on any local SKILL.md.

Video coming soon — see the install commands below to try it now.

Advanced — run the commands yourself

$ curl -fsSL https://raw.githubusercontent.com/qingxuantang/tar-engine/master/setup-mcp.sh -o setup-mcp.sh
$ chmod +x setup-mcp.sh
$ ./setup-mcp.sh                   # Claude Code (default); add --client cursor|codex
$ # in Claude Code:
$ /mcp list   # tar-engine should appear

Read the MCP integration guide → Self-host with Docker →

Want this without running Docker?

Hosted TAR Engine MCP is coming. Drop your email and we'll tell you when it's live.