The Agent Skills Directory

[PROMPT_INJECTION]: The skill uses detailed persona instructions and activation keywords to define the agent's character and workflow. These are designed for specific role-play and task handling rather than bypassing security guardrails or extracting system prompts.\n- [DATA_EXFILTRATION]: The skill interacts with project management data via internal tools. It explicitly mandates a confirmation step (Y2 protocol) before any data is written to external systems like Yunxiao, ensuring user oversight and preventing unauthorized data flow.\n- [COMMAND_EXECUTION]: The skill utilizes helper tools (e.g., $oneos-autoprd, $YunxiaoPMapp) to automate development tasks. These platform-integrated tools are used for their intended purpose of generating requirements and tracking project status.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest external materials for summarization. It mitigates potential injection risks by implementing semantic evaluation protocols (G1/G2 gates) to verify the rationality of requirements before they are processed.\n- [SAFE]: The skill's operations, including local file access and tool orchestration, are consistent with its stated purpose as a development and project management assistant. No malicious patterns, obfuscation, or unauthorized external communications were detected.

wangmian-twin