The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/archive.py utilizes subprocess.run to interact with the Obsidian CLI for managing file properties and moving project folders. This is implemented safely by passing arguments as a list rather than a single shell string, which prevents command injection. Furthermore, the script performs explicit path validation to ensure that any target files are located within the current Obsidian vault before execution.
[DATA_EXFILTRATION]: No malicious network operations or credential harvesting patterns were detected. While the skill performs research and SEO analysis, these actions are performed by the agent using standard tools or by the user manually, and the resulting data is processed locally by scripts without being transmitted to unauthorized external servers.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from user-supplied 'raw material' and external 'seeds'. However, the workflow is designed with significant safeguards, including a mandatory planning stage where the agent generates a structured 'Plan' for user review before drafting the final article. This human-in-the-loop checkpoint, combined with the fact that the processing scripts have restricted file-system access, effectively mitigates the risk of indirect prompt injection attacks.

wechat-writer