security-alert-monitor

HIGH W007: Insecure credential handling detected in skill instructions.

• Insecure credential handling detected (high risk: 0.90). The skill requires the agent to quote or paraphrase email evidence from scanned threads (and produce that text in its output) which can include credentials or secrets found in emails, and it does not instruct redaction—creating a high risk of verbatim secret exfiltration.

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

• Third-party content exposure detected (high risk: 0.85). The required workflow calls an MCP “search” tool to retrieve email threads (outsider-authored content from other people’s messages in the user’s mailbox) and then passes the retrieved thread text into the agent via the subsequent “ask” step for LLM analysis.

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

• Potentially malicious external URL detected (high risk: 0.90). The skill explicitly requires the iGPT MCP at https://mcp.igpt.ai/ for runtime tool calls (search/ask), meaning it depends on that external service to execute tooling and drive prompts/behavior during execution.