alterlab-deep-research
Pass
Audited by Gen Agent Trust Hub on May 5, 2026
Risk Level: SAFE
Full Analysis
- [PROMPT_INJECTION]: No malicious override or bypass patterns were detected. Instructional language (e.g., 'IMPORTANT', 'Critical') is used naturally within the context of academic rigor and process control. There are no attempts to extract system prompts or bypass safety filters.
- [DATA_EXFILTRATION]: The skill does not access sensitive local files (e.g., .ssh, .aws) or hardcode credentials. It performs network operations limited to academic databases and metadata resolution (e.g., DOI resolution, Google Scholar), which are well-known and trusted services.
- [REMOTE_CODE_EXECUTION]: No executable shell scripts or Python code targeting remote servers were found. Software mentions (e.g., R and Python packages for meta-analysis) are provided as references for researchers to implement manually, rather than for direct agent execution.
- [OBFUSCATION]: The content is presented in clear, plain-text Markdown and Traditional Chinese. No Base64, zero-width characters, homoglyphs, or other obfuscation techniques were identified.
- [SAFE]: The skill demonstrates best practices in academic research security. It includes a mandatory 'Ethics Review Agent' and 'Source Verification Agent' that systematically check reference integrity (50% verification threshold), predatory journal indicators, and potential conflicts of interest. The use of a 'Devil's Advocate' agent further mitigates cognitive and research design biases.
Audit Metadata