Security audit
Security checks for vulnerabilities and agentic risk
This looks like a startup idea validation helper, but it asks for broad shell-command access and can write workspace files without clear scoping or confirmation.
Review this skill before installing. It is not showing malware or hidden exfiltration, but it grants shell execution for a workflow that mostly needs research and text generation. Use it only where command execution and docs/prd.md creation are acceptable, and require confirmation before running shell commands or writing/replacing files.
The skill is presented as a lightweight idea-validation tool, but the manifest and instructions extend into generic command execution and network-diagnostic behavior. That mismatch broadens the effective capability of the skill beyond user expectations, increasing the chance an agent will invoke filesystem or shell actions unrelated to validation without sufficient scrutiny.
Including shell or network-diagnostic execution capability in a validation skill creates an unnecessary attack surface. Even if no explicit exploit payload is embedded, agents may treat these instructions as permission to run commands such as ping or other diagnostics, which can lead to policy bypass, environmental probing, or misuse of the host context.
Overly broad activation keywords can cause the skill to trigger in loosely related startup conversations, increasing the chance of unintended tool use or file operations in contexts the user did not mean to invoke this skill. In an agent environment, weak scope boundaries are a security concern because they widen when privileged behaviors become active.
The skill instructs generation of docs/prd.md without warning that it will write to the workspace. Silent or implicit file creation is dangerous in agent settings because users may believe the tool is advisory-only, while the agent is actually modifying project files and potentially overwriting existing content.
Sections for Manifest对齐, 魔鬼代言人, STREAM分析, 单位经济, and 死亡初创搜索 repeatedly state '不支持', but immediately follow with text saying to parse their input parameters, complete core logic, and return structured responses. That is an active contradiction in the file's stated intent, not just an omission.
No suspicious patterns detected.
