The Agent Skills Directory

[SAFE]: The skill performs local data processing on research paper metadata. Analysis of the scripts and configuration files confirms that the skill does not attempt network exfiltration, access sensitive system credentials, or employ obfuscation techniques.- [COMMAND_EXECUTION]: Local script execution is restricted to the provided Python implementation for paper deduplication and ranking. The logic is self-contained and does not use dangerous functions like eval() or subprocess calls for arbitrary command execution.- [PROMPT_INJECTION]: The skill processes untrusted external metadata (paper titles and abstracts). While this represents a surface for indirect prompt injection if the output is consumed by an LLM, the current skill processes this data deterministically without interpolating it into agent prompts.- [DATA_EXFILTRATION]: No network activity was detected. The script reads input from local files and writes the processed core set to the workspace without making external requests.

dedupe-rank