The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute multiple Python scripts (e.g., search_arxiv.py, scan_existing_notes.py, link_keywords.py) located in a specific host directory server/scripts/research-news/. These scripts are not included in the skill package, and their logic cannot be verified.
[DATA_EXFILTRATION]: The workflow requires the agent to scan a directory path defined by $VAULT_PATH to build keyword indexes. This represents a data exposure risk as it accesses personal note repositories (common in Obsidian or Logseq workflows) using unverified local scripts.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes untrusted content from external sources (arXiv abstracts, social media posts from X and Xiaohongshu).
Ingestion points: Data retrieved from arXiv API, Semantic Scholar API, X (Twitter), and Xiaohongshu via unverified scripts.
Boundary markers: No delimiters or instructions are provided to the agent to ignore or sanitize embedded instructions within the ingested paper abstracts or social media content.
Capability inventory: The skill uses python command execution and the write_file tool to generate outputs and linked documents.
Sanitization: No validation or sanitization steps are defined for the data entering the recommendation pipeline.
[EXTERNAL_DOWNLOADS]: The skill mentions a source repository at https://github.com/evil-read-arxiv. This repository and organization are not recognized as trusted or well-known services.

research-news