The Agent Skills Directory

[DATA_EXPOSURE_AND_EXFILTRATION]: The skill instructions in references/delivery-rules-chapter.md direct the agent to read local files within the ~/Desktop/ONE-OS/work/CURSOR/ directory. This is used to retrieve the original requirements for comparison and verification. The access is limited to a specific project subfolder, and there is no evidence of data being sent to external servers.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from chat logs and meeting transcripts. To mitigate risks, the skill includes explicit processing and sanitization rules in references/rules.md, such as removing conversational filler and adhering to a formal writing style. The agent's capabilities are restricted to text generation and interaction with the $YunxiaoPM tool, minimizing the impact of potential injection attacks.

AutoRDO