The Agent Skills Directory

[PROMPT_INJECTION]: The skill strictly defines a persona and interaction guidelines. It includes internal safety constraints, such as a refusal to provide specific stock recommendations ("我不荐股，这是底线"). No instructions to bypass AI safety filters or reveal system prompts were found.
[DATA_EXFILTRATION]: No sensitive file access or unauthorized network operations were detected. The skill uses a standard web search tool for market research as part of its "Agentic Protocol," which is standard behavior for an investment analysis assistant.
[REMOTE_CODE_EXECUTION]: The skill consists entirely of Markdown instructions and reference data. There are no scripts, binaries, or commands that attempt to download and execute code from remote sources.
[INDIRECT_PROMPT_INJECTION]: The skill ingests user queries and performs web searches. While it processes untrusted external data (search results), its capabilities are limited to generating conversational responses. It does not have the ability to write files or execute shell commands based on the processed data.
[COMMAND_EXECUTION]: No usage of dangerous shell commands or privilege escalation attempts were found. The installation instruction provided in the documentation uses the standard platform command.

zettaranc-perspective