The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructions and scripts reference the installation of the playwright library and the chromium browser engine to support browser automation tasks.
[COMMAND_EXECUTION]: The skill utilizes a local Python script scripts/x_post.py which executes shell commands to control a Playwright browser instance for interacting with X.com.
[PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it ingests untrusted article drafts provided by the user and has the capability to perform write operations (posting or saving drafts) to an external service.
Ingestion points: Article drafts provided as text or file paths (SKILL.md).
Boundary markers: The instructions include rules to preserve code and links, but do not explicitly define delimiters for the input draft text to prevent the AI from following embedded instructions.
Capability inventory: The script scripts/x_post.py allows for saving drafts and publishing posts on X.com.
Sanitization: No specific sanitization or filtering of the input draft content is mentioned.
Mitigation: The risk is significantly reduced by a mandatory human-in-the-loop requirement; the agent is instructed to only publish (send) when the user has provided a verbatim final文案 and explicit confirmation via a --yes flag.

soia-media-publish-x-thread