The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes user-supplied business困境 (business dilemmas) and retrieves external case data that may contain untrusted content. \n
Ingestion points: User descriptions of business scenarios and case records retrieved from local storage or the vendor's remote API at api.xyskill.xyz. \n
Boundary markers: While the skill specifies a structured Phase 3 report format, it lacks explicit delimiters or instructions to ignore potentially malicious embedded commands within the retrieved case data. \n
Capability inventory: The agent can execute a Python script, read local files, and communicate with a remote API. The agent is instructed to incorporate the retrieved knowledge directly into its final analysis and recommendations. \n
Sanitization: No specific sanitization or filtering logic is mentioned to prevent data-embedded instructions from influencing the agent's behavior. \n- [COMMAND_EXECUTION]: The skill facilitates the execution of a bundled Python helper script (atoms-search.py) to perform business case lookups. \n
The agent is instructed to run this script using python3 to query the business case database. This represents a technical capability for shell interaction, where arguments are derived from user-provided keywords. \n- [EXTERNAL_DOWNLOADS]: The atoms-search.py script performs network operations to fetch business case data from a remote server. \n
The script communicates with https://api.xyskill.xyz/v1/atoms/search to provide extended case search capabilities. This network activity is a functional part of the skill's design, used to access the vendor's centralized business case database (author: xyaz1313).

xy-precedent