The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/wechat_egress.py uses subprocess.Popen to execute the ssh command with arguments -N -D to establish a SOCKS5 proxy tunnel. This is intended to ensure a fixed outbound IP address for WeChat API requests, as documented in the skill's egress configuration.
[COMMAND_EXECUTION]: The script scripts/wechat_draft.py uses subprocess.run to invoke prepare_article.py for validating the generated HTML content against the source Markdown.
[DYNAMIC_EXECUTION]: The script scripts/wechat_images.py utilizes __import__("sys") for error reporting to stderr.
[INDIRECT_PROMPT_INJECTION]: The skill processes user-provided Markdown content and converts it into HTML for platform delivery, which presents an attack surface for indirect prompt injection if the content contains malicious instructions designed to be followed during the agent's verification phase.
Ingestion points: Markdown articles are read as input in scripts/render_article.py and validated in scripts/prepare_article.py.
Boundary markers: The skill instructions emphasize working with finalized articles ("定稿") and include a mandatory "回读核对" (back-reading verification) step.
Capability inventory: The skill can perform network operations (WeChat API), file system writes (image caching, frontmatter updates), and subprocess execution (SSH, validation scripts).
Sanitization: scripts/prepare_article.py implements a filter for forbidden HTML elements (e.g., script, style, div) and attributes (e.g., class, id) to ensure the output remains a clean WeChat-compatible fragment.

ray-wechat