grep-mcp

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 1.00). The skill explicitly instructs the agent to ask the user to paste a long-lived "parcha-xxx" API key and to insert that key verbatim into .mcp.json and into curl/Authorization headers, which requires the LLM to handle and output secret values directly.

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

• Third-party content exposure detected (medium risk: 0.30). Grep MCP tools ingest user-provided research_create.question via POST /api/v2/research (and then return results via research_get/file read), so outsider-authored free text can be submitted into the workflow through the model’s tool calls without pre-selecting a specific trusted item.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill configures and calls the Grep MCP endpoint (e.g., https://api.grep.ai/api/v2/mcp) at runtime (via .mcp.json and a curl verification) which exposes tools the model can invoke to execute remote actions.