The Agent Skills Directory

[PROMPT_INJECTION]: No evidence of prompt injection or behavior override instructions was found. The skill maintains a highly professional academic register and focuses on quality control rather than bypassing constraints.
[DATA_EXPOSURE_AND_EXFILTRATION]: The skill demonstrates high awareness of data privacy. It includes an 'Anti-Leakage Protocol' (references/anti_leakage_protocol.md) specifically designed to isolate provided research materials from the model's general knowledge and prevent data leaks. No sensitive file paths, hardcoded credentials, or unauthorized network operations were detected.
[OBFUSCATION]: All analyzed content is provided in clear, standard Markdown and LaTeX templates. No Base64, homoglyphs, or zero-width character techniques were detected.
[DYNAMIC_EXECUTION]: The 'visualization_agent' is designed to generate Python (matplotlib/seaborn) and R (ggplot2) code for rendering academic charts. While this involves dynamic code generation, the instructions are scoped to statistical visualization using standard libraries and include mandatory safety/quality checks. The code is intended for user-guided rendering of research results, which is appropriate for the skill's stated purpose.
[INDIRECT_PROMPT_INJECTION]: The skill has a defined surface for ingesting user data (past papers for style calibration, research notes, reviewer comments). However, it implements a 'Knowledge Isolation Directive' that explicitly instructs the agent to treat this data as factual content rather than instructions, mitigating the risk of indirect injection.
[EXTERNAL_DOWNLOADS]: All external URLs referenced in the venue disclosure policies (e.g., Nature, Science, ACL) point to official, well-known academic domains for the purpose of policy lookup. No suspicious or risky downloads were identified.

academic-paper