The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill generates shell scripts (based on record_template.sh) and Node.js one-liners by interpolating user-provided data directly into code strings. For instance, the browser_goto function in record_template.sh uses node -e to execute a script where the $url variable is placed inside a string literal. If a malicious URL containing single quotes and JavaScript commands is provided, it could lead to arbitrary code execution within the Node.js process.
[INDIRECT_PROMPT_INJECTION]: The workflow relies on extracting commands from user-supplied "voice-over scripts" (script.md) using simple quote detection. This ingestion process lacks boundary markers or sanitization, potentially allowing an attacker to embed malicious commands within a script that the skill will then dutifully include in the generated send_cmd sequence for execution in a claude CLI session.
[COMMAND_EXECUTION]: The skill facilitates the automated execution of CLI commands via tmux send-keys and browser actions via Playwright. While this is the primary functionality, the lack of validation on the generated command sequence poses a risk if the input source is untrusted.

laohan-luping