ml-paper-writing

[Skill Scanner] Instruction to copy/paste content into terminal detected All findings: [CRITICAL] command_injection: Instruction to copy/paste content into terminal detected (CI012) [AITech 9.1.4] [HIGH] autonomy_abuse: Skill defers behavior to remote content fetched at runtime (AU005) [AITech 1.2] This skill is primarily benign and coherent with its stated purpose of assisting ML paper writing: repository inspection, LaTeX templating, and programmatic citation verification are appropriate. However, the explicit recommendation to install and route queries through a third-party MCP (https://mcp.exa.ai/mcp) using npx (remote package execution) is a supply-chain and data-leakage risk. That download-and-execute pattern and routing of potentially sensitive search queries/metadata through a non-official proxy is disproportionate and suspicious. If you plan to use this skill, avoid installing or configuring the Exa MCP remote server unless you fully trust and have audited that service; prefer direct calls to Semantic Scholar, arXiv, and CrossRef. No clear malware or obfuscated code found in the text itself. LLM verification: This SKILL.md is primarily an instructional skill to assist with writing ML papers; its capabilities align with its purpose. It contains no explicit malicious code, hardcoded secrets, or obfuscated payloads. However, it includes supply-chain and data-exposure risks: recommending an unpinned npx installation pointing to a custom MCP endpoint (mcp.exa.ai) and multiple copy/paste shell commands which could execute remote code or leak unpublished repository content. That pattern is a legitimate supp