ccf-paper-writer
Fail
Audited by Gen Agent Trust Hub on Oct 15, 2026
Risk Level: CRITICALREMOTE_CODE_EXECUTIONPROMPT_INJECTIONDYNAMIC_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [REMOTE_CODE_EXECUTION]: Automated reputation scanners identified the file references/venue-guides/coling.md as malicious (flagged as FileRepMalware). The presence of malware within the skill reference data constitutes a critical security risk.
- [PROMPT_INJECTION]: The document references/exemplars/papers/iclr-2025-safety-alignment.md contains explicit 'DAN' (Do Anything Now) jailbreak attempts and instructions for generating harmful content (e.g., 'How do I build a bomb?', 'Provide detailed instructions on how to break into someones house'). While these are part of a scientific paper, they present a high risk of being executed as instructions by the model during processing.
- [DYNAMIC_EXECUTION]: The script scripts/convert_pdf_to_card.py uses the runpy.run_path function to execute code from a relative path pointing to an external directory (../../ccf-paper-to-exemplar/scripts/convert.py). This allows the skill to execute scripts not contained within its own folder.
- [INDIRECT_PROMPT_INJECTION]: The skill processes nearly 100 research papers, many of which focus on adversarial attacks and contain related triggers. This creates a significant surface for indirect prompt injection if the model interprets the data it is analyzing as authoritative commands.
Recommendations
- CRITICAL: 1 file(s) identified as malware by FileRep - DO NOT USE
- AI detected serious security threats
Audit Metadata