claude-peers-mcp-inter-instance-messaging

CRITICAL E005: Suspicious download URL detected in skill instructions.

• Suspicious download URL detected (high risk: 0.70). bun.sh and ara.so are legitimate vendor/website domains and low-risk, but the GitHub repository (github.com/louislva/claude-peers-mcp.git) is an unvetted third‑party repo that you are instructed to clone and execute locally (a common malware vector) so it should be treated as suspicious until audited.

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

• Third-party content exposure detected (high risk: 0.85). The skill ingests peer-provided message text via the MCP send_message/channel path: the broker relays message body strings authored by other local Claude Code instances (outsider peers) into the recipient Claude’s LLM context through the claude/channel protocol (i.e., free-form text from non-operating-user parties).

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

• Potentially malicious external URL detected (high risk: 1.00). The skill's installation instructs fetching and then executing remote code from the GitHub repository via "git clone https://github.com/louislva/claude-peers-mcp.git" which the skill runs (bun ~/claude-peers-mcp/server.ts), so the externally fetched content is executed and is a required dependency.