claude-plugin-auditor

# Plugin Security Auditor Audit Claude Code plugins and MCP servers for security risks before installing them. **36.7% of MCP servers have SSRF vulnerabilities. 43% have command injection paths. 41% have zero authentication.** This plugin helps you check before you trust. ## What it does Scans plugin files looking for: - **Exfiltration** — `fetch`/`http` calls to external URLs, sending `process.env` or system paths - **Secret reading** — accessing `.env`, `~/.ssh`, `~/.aws`, `~/.claude`, credentials - **Hidden execution** — `eval()`, `child_process`, dynamic `require()`/`import()` - **Obfuscation** — minified code, base64-encoded commands, meaningless variable names - **Supply chain** — `npx @latest` patterns, unpinned versions, runtime downloads - **Prompt injection** — hidden malicious instructions in `.md` files - **Excessive permissions** — capabilities beyond what the plugin claims to need ## Install Copy this plugin to your Claude Code plugins directory: ```bash git clone https://github.com/0001-YIDev/claude-plugin-auditor ~/.claude/plugins/plugin-security-auditor ``` Or install from the marketplace: ``` /plugin install plugin-security-auditor ``` ## Usage ``` /audit security-guidance /audit ~/.claude/plugins/some-plugin/ /audit https://github.com/user/mcp-server ``` ## Verdicts | Verdict | Meaning | |---------|---------| | **SAFE** | Only `.md` instructions, no executable code | | **SAFE+CODE** | Has JS/TS but transparent, no suspicious access | | **REVIEW** | Ambiguous patterns, needs human review | | **DO NOT INSTALL** | Clear exfiltration, secret reading, or hidden execution | ## Example output ``` ## Audit: example-plugin **Source:** marketplace official **Author:** SomeCompany **Files analyzed:** 4 total (2 .md, 1 .js, 1 .json) **Findings:** - [OK] plugin.json — clean metadata, no suspicious URLs - [RISK] hooks/main.js:23 — fetch() to external URL not matching declared service - [OK] README.md — no prompt injection **What it actually does:** Hooks into Edit tool to send file paths to analytics endpoint **Verdict:** REVIEW **Recommendation:** Review hooks/main.js line 23 manually ``` ## Why this exists The Claude Code plugin ecosystem is growing fast (11,000+ MCP servers, 232% growth in 6 months) but security tooling hasn't kept up: - [30 CVEs in 60 days](https://www.heyuan110.com/posts/ai/2026-03-10-mcp-security-2026/) in Q1 2026 - [GitHub MCP exploited](https://invariantlabs.ai/blog/mcp-github-vulnerability) to access private repos - 3 CVEs in Anthropic's own MCP Git server (including RCE) - 5 supply chain incidents in 2 weeks Enterprise tools exist (Runlayer $11M, Levo.ai, MintMCP) but nothing for individual developers. This plugin fills that gap. ## License MIT