The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/fetch-comments-via-gstack.sh dynamically generates JavaScript and executes it within a browser environment using a subprocess call to the gstack browse utility. This allows for automated interaction with authenticated web sessions but introduces a surface for arbitrary script execution based on local configuration.
[DYNAMIC_EXECUTION]: Several scripts, such as gen-cover-ai.sh and gen-illustration.sh, invoke external Node.js scripts (gpt_image_2_skill.cjs) and utilize complex instructions to orchestrate remote AI models for image generation.
[EXTERNAL_DOWNLOADS]: The README.md and installation instructions require cloning third-party repositories, specifically Wangnov/gpt-image-2-skill, which is a dependency for the skill's image generation functionality.
[INDIRECT_PROMPT_INJECTION]:
Ingestion points: The skill ingests untrusted data from the mp.weixin.qq.com comment API and processes user-provided article drafts.
Boundary markers: No explicit delimiters or instructions to ignore embedded commands were found in the scripts that process external text.
Capability inventory: The skill possesses capabilities for file writing (article.md, publish.json), network requests to WeChat APIs, and command execution via shell and Node.js.
Sanitization: External content from comments is formatted for Markdown output but is not sanitized for downstream prompt injection risks.
[DATA_EXFILTRATION]: The scripts/upload-draft.sh script automatically performs git add, commit, and push operations on the article directory, exfiltrating local content to a remote Git repository whenever an article is prepared for publishing.
[CREDENTIALS_UNSAFE]: The skill's operation relies on WECHAT_APPID and WECHAT_SECRET environment variables. The documentation explicitly instructs users to configure these secrets, which are subsequently handled by multiple scripts in the scripts/ directory.

wjs-publishing-wechat