The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions do not contain attempts to override agent behavior, bypass safety guidelines, or extract system prompts. The logic is strictly focused on document analysis and presentation structure.
[DATA_EXPOSURE_AND_EXFILTRATION]: No sensitive file paths, hardcoded credentials, or unauthorized network operations were detected. The skill operates on local document inputs and generates local outputs.
[OBFUSCATION]: Analysis revealed no Base64 encoded commands, zero-width characters, homoglyphs, or hidden strings designed to conceal malicious intent.
[REMOTE_CODE_EXECUTION]: No patterns of remote script execution (e.g., curl|bash) or dynamic evaluation of untrusted input were found. Execution is restricted to defined Python logic.
[UNVERIFIABLE_DEPENDENCIES]: The skill relies on standard, industry-recognized libraries (PyMuPDF, Pillow, and python-pptx) for its core functionality. It includes a policy for local dependency management that follows standard development practices.
[INDIRECT_PROMPT_INJECTION]: While the skill processes untrusted external data (scientific papers), this is the primary purpose of the tool. The instructions include structured workflows and quality assurance steps that mitigate the risk of accidental instruction obedience from paper content.
[PRIVILEGE_ESCALATION]: The instructions do not use sudo, modify system-level configurations, or attempt to gain elevated permissions.

nature-paper2ppt