The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill requires the user to clone several external repositories to function, including the author's own tools (md2wechat, gstack) and a third-party image generation utility (gpt-image-2-skill).
[REMOTE_CODE_EXECUTION]: The skill's scripts execute code from these external dependencies at runtime. For instance, the image generation scripts (gen-cover-ai.sh and gen-illustration.sh) invoke Node.js scripts located within the cloned gpt-image-2-skill repository.
[DYNAMIC_EXECUTION]: The skill employs dynamic execution techniques to manage browser sessions and fetch data. This includes using AppleScript (in open-draft-edit.sh) to programmatically control Safari or Chrome, and using a browser wrapper (in fetch-comments-via-gstack.sh) to execute dynamically assembled JavaScript for fetching comments using existing cookies.
[COMMAND_EXECUTION]: Several core scripts (such as upload-draft.sh, mass-send.sh, and fetch-comments.sh) execute complex logic by piping Python and Node.js code through shell heredocs, which can be an avenue for command injection if inputs are not properly handled.
[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted user-supplied text (article drafts and notes) and interpolates it into prompts for image generation tools. While it uses templates, it lacks robust delimiters to prevent embedded instructions from affecting the image generation process.
Ingestion points: article.md (containing user-provided drafts and notes).
Boundary markers: Uses template placeholders like [目标字词] and [文章内容] but lacks specific boundary delimiters to isolate user text from instructions.
Capability inventory: Network communication (WeChat API, Git), file system modifications (writing articles and metadata), and subprocess execution (image generation and browser control).
Sanitization: The skill includes a 'Privacy Scanning' step (defined in SKILL.md and STYLE.md) designed to detect and remove sensitive personal data (names, addresses, phone numbers) before publication, which serves as a mitigation for data exposure.

wjs-publishing-wechat