The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: In scripts/web_search.py, the skill uses uvx to dynamically download and execute the minimax-coding-plan-mcp package at runtime to provide search functionality.
[COMMAND_EXECUTION]: The skill executes shell commands to initialize the MCP server using StdioServerParameters in the web_search.py script.
[PROMPT_INJECTION]: The skill is vulnerable to Indirect Prompt Injection (Category 8) as it retrieves data from the internet and injects it into the agent's context.
Ingestion points: scripts/web_search.py ingests untrusted search results (titles, links, and snippets) from the web via the MiniMax tool.
Boundary markers: Absent. The search results are returned as raw text or JSON structures without clear delimiters to prevent the agent from obeying instructions found in search results.
Capability inventory: The agent uses these results to answer user questions, potentially exposing it to malicious instructions embedded in third-party websites.
Sanitization: No filtering or sanitization of the retrieved content is performed before returning it to the agent.
[DATA_EXFILTRATION]: The skill transmits user queries to https://api.minimaxi.com. This is the official endpoint for the service described and is considered a well-known service operation.

minimax-web-search