The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The scripts/publish.sh script automatically installs the @wenyan-md/cli package globally using npm install -g if it is not present on the system. This introduces a supply chain risk through the execution of unpinned code from an external registry.
[CREDENTIALS_UNSAFE]: The skill reads sensitive API credentials from $HOME/.openclaw/workspace/TOOLS.md. The scripts scripts/publish.sh and scripts/setup.sh use shell commands to extract WECHAT_APP_ID and WECHAT_APP_SECRET. Direct reading of credential files via script is a sensitive operation.
[COMMAND_EXECUTION]: The skill executes shell commands including the wenyan publishing tool, npm, and text processing utilities. It passes user-controlled file paths as arguments, which can be a vector for command misuse.
[PROMPT_INJECTION]: The skill processes untrusted Markdown data as input for the publishing pipeline, creating an indirect injection surface.
Ingestion points: Markdown files processed by scripts/publish.sh.
Boundary markers: No delimiters or warnings are used to isolate embedded instructions in the Markdown content.
Capability inventory: Command execution (npm install, wenyan publish) and local file reading (TOOLS.md).
Sanitization: No validation or filtering of Markdown content is performed before processing.

wechat-publisher