security-scanner

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The prompt explicitly tells WebFetch to "Return the raw file content exactly as-is" and instructs reading local and repo files (e.g., .mcp.json, installed plugins, SKILL.md), which forces the LLM to ingest — and potentially include — secret values verbatim, creating a high exfiltration risk.

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.90). Although many entries are common GitHub URL patterns (generally low-risk by themselves), the list explicitly includes direct malicious endpoints (e.g., https://evil.com/payload and https://attacker.com/collect) and raw/download endpoints that can be used to serve executables or scripts, so taken together this set is highly suspicious as a malware distribution vector.

MEDIUM W011: Third-party content exposure detected (indirect prompt injection risk).

• Third-party content exposure detected (high risk: 0.90). The skill explicitly fetches and reads arbitrary public GitHub repositories and files via the --url GitHub support (using api.github.com and raw.githubusercontent.com) and also ingests SKILL.md/agent prompt files from user-installed skill/plugin directories, meaning it consumes untrusted third‑party user-generated content that the agent will read and interpret.

MEDIUM W012: Unverifiable external dependency detected (runtime URL that controls agent).

• Potentially malicious external URL detected (high risk: 0.90). The skill performs runtime fetches from GitHub (e.g. https://raw.githubusercontent.com/{owner}/{repo}/{branch}/{path} and https://api.github.com/repos/{owner}/{repo}/contents/{path}?ref={branch}) to retrieve SKILL.md and agents/*.md files which contain agent system prompts/instructions and are required inputs for the URL-based scan, so the fetched content can directly control prompts.