The Agent Skills Directory

[COMMAND_EXECUTION]: The skill interpolates user-provided file paths and directory names directly into shell commands (e.g., node "$PLUGIN_DIR/scripts/extract_mermaid.js" {file}). This allows for command injection if the user provides a path containing shell metacharacters like semicolons or pipe symbols.
[REMOTE_CODE_EXECUTION]: The skill uses find to dynamically resolve its script directory ($PLUGIN_DIR) and then executes scripts located within that directory (e.g., bash "$PLUGIN_DIR/scripts/ensure-deps.sh"). This discovery includes the user's entire home directory, allowing an attacker to place a malicious script in a matching path to achieve code execution.
[COMMAND_EXECUTION]: The skill constructs shell commands using variables from local configuration files (.claude/mermaid.json) and user input. The lack of explicit sanitization before passing these strings to the Bash tool creates multiple injection surfaces.

mermaid-render