The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructions in SKILL.md direct the agent to execute a local Python utility using a shell command pattern: python3 <本 skill 目录>/scripts/atoms-search.py "<关键词>" --skill xy-ops -k 5.
User-provided input is interpolated into the <关键词> placeholder. This creates a command injection surface if the agent does not properly sanitize the input (e.g., escaping backticks or subshells) before passing it to the operating system's shell.
[EXTERNAL_DOWNLOADS]: The scripts/atoms-search.py script makes outbound network requests to a remote API at https://api.xyskill.xyz/v1/atoms/search using the urllib.request library.
This resource is associated with the skill's vendor infrastructure and is used to fetch dynamic business intelligence data (atoms) at runtime.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection due to its design of processing untrusted data and routing it to executable capabilities.
Ingestion points: The skill solicitations user business states and specific questions in Phase 1 and Phase 2, which are then used to formulate search queries.
Boundary markers: The skill employs specific markdown table structures and a custom "stop rule" in Phase 2 to regulate output, though these do not prevent adversarial data from reaching the execution layer.
Capability inventory: The skill possesses the ability to execute shell commands and perform network operations via the included atoms-search.py script.
Sanitization: The skill relies on the agent's internal safety filters rather than explicit instruction-level sanitization to prevent malicious user input from being executed as part of the command line string.

xy-ops