aris-auto-paper-improvement-loop
Pass
Audited by Gen Agent Trust Hub on May 6, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTIONDATA_EXFILTRATION
Full Analysis
- [PROMPT_INJECTION]: The skill ingests untrusted LaTeX source files from a user-provided directory and sends them to an external LLM for review. This creates an indirect prompt injection surface where malicious instructions hidden in the LaTeX content could influence the reviewer LLM to suggest dangerous "fixes" that the agent then implements using powerful system tools.- [COMMAND_EXECUTION]: The skill utilizes the
Bash(*)tool to perform complex file operations, recompile LaTeX papers, and implement changes based on LLM output. The autonomous execution of commands derived from external LLM responses and user-controlled directory paths presents a risk of command injection or unintended system state changes.- [DATA_EXFILTRATION]: The skill reads from~/.claude/feishu.jsonto retrieve configuration for sending notifications to the Feishu platform. Accessing sensitive configuration files in the agent's internal directory to facilitate external network communication represents a potential data exposure and exfiltration risk.
Audit Metadata