The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions direct the agent to execute a bundled Python script using user-supplied keywords: python3 <本 skill 目录>/scripts/atoms-search.py "<议题关键词>" -k 5. This pattern allows user input to be interpolated directly into a shell command, which may lead to command injection if the agent does not properly sanitize the input.
[INDIRECT_PROMPT_INJECTION]: The skill ingests "atoms" (knowledge units) retrieved from a remote vendor API (api.xyskill.xyz) or a local JSONL file to generate agent responses. There are no boundary markers or instructions provided to distinguish this external data from system instructions.
Ingestion points: scripts/atoms-search.py (retrieves data from API or references/atoms.jsonl).
Boundary markers: None identified in the response generation templates.
Capability inventory: Shell command execution via atoms-search.py and file writing to ./xy-深度分析/ for reports.
Sanitization: The skill does not define validation or escaping for the retrieved knowledge units before they are processed by the LLM.
[EXTERNAL_DOWNLOADS]: The atoms-search.py script makes network requests to https://api.xyskill.xyz/v1/atoms/search to fetch knowledge data. This domain is a vendor-owned resource belonging to the author xyaz1313.
[DATA_EXPOSURE]: The bundled script scripts/atoms-search.py searches for a configuration file at ~/.xy/config.json. Accessing hidden files in the user's home directory is a sensitive operation, although it appears to be part of the vendor's configuration system.

xy-roundtable