The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/archive.py utilizes subprocess.run to interact with the Obsidian CLI for setting file properties. This execution is performed using a list-based argument structure rather than a shell string, which significantly reduces the risk of shell command injection. The tool being called is a standard productivity application, and the operations are restricted to metadata management within the user's workspace.
[INDIRECT_PROMPT_INJECTION]: The skill processes data from external sources, such as a 'clip' directory for web clippings and user-provided raw materials, which represents a potential attack surface for indirect prompt injection. However, the skill explicitly mitigates this risk through its 'Human-AI Boundary' (人机边界卡) and 'Topic Gate' instructions. These require the agent to strictly distinguish between verified author facts and untrusted external inputs, providing a clear boundary for the model's reasoning.
[SAFE]: The skill implements several proactive security checks, such as verifying that archive operations occur only within the designated articles/Project_* directories, preventing path traversal or unauthorized file modifications. Additionally, all scripts are implemented using Python's standard library, avoiding unverifiable third-party dependencies.

wechat-writer