The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/full-page-screenshot.mjs is vulnerable to command injection. It uses child_process.execSync to run sips and python3 with file paths constructed from the outputFile positional argument. Since this argument is not sanitized, a malicious path containing shell metacharacters could result in arbitrary command execution on the host.\n- [DYNAMIC_EXECUTION]: The tool employs dynamic execution by generating a temporary Python script (_stitch.py) at runtime to merge image tiles for very tall pages. It also uses the Chrome DevTools Protocol's Runtime.evaluate method to execute complex, multi-line JavaScript logic directly within the context of the target web page to manipulate the DOM and manage lazy-loading.\n- [EXTERNAL_DOWNLOADS]: The skill's documentation directs users to install an external Python dependency, Pillow, using pip3 install Pillow to support the image stitching functionality for pages exceeding 16,000 pixels.\n- [INDIRECT_PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection as it processes content from untrusted external web pages and uses that content to influence its internal logic.\n
Ingestion points: Web page structure and metadata accessed through the Chrome DevTools Protocol in scripts/full-page-screenshot.mjs.\n
Boundary markers: None; the script executes code and reads data directly from the untrusted page environment.\n
Capability inventory: File system write operations (fs.writeFileSync), network communication via WebSockets, and shell command execution (execSync).\n
Sanitization: There is no evidence of sanitization or validation of the data retrieved from the web page (e.g., element counts, scroll heights) before it is used to control the script's execution flow.

full-page-screenshot