The Agent Skills Directory

[COMMAND_EXECUTION]: The run.js script allows the execution of arbitrary Javascript code provided as a command-line argument or via stdin. It writes this input to a temporary file and runs it with the Node.js interpreter, providing unrestricted access to host resources through modules like fs and child_process.
[REMOTE_CODE_EXECUTION]: The skill facilitates remote code execution by allowing the agent to run dynamically generated scripts and automatically downloading necessary binaries via npm install and npx playwright install at runtime.
[EXTERNAL_DOWNLOADS]: The installPlaywright function in run.js triggers npm install and npx playwright install to fetch the Playwright library and browser binaries from official registries.
[PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection via its web scraping capabilities.
Ingestion points: Functions in lib/helpers.js such as getPageText, getPageStructure, and describePageForAI retrieve data from arbitrary external URLs.
Boundary markers: None present; content is returned as raw strings or objects.
Capability inventory: The skill has high-privilege system access via the run.js code execution bridge.
Sanitization: No validation or filtering is performed on data retrieved from external sources before processing.
[COMMAND_EXECUTION]: The launchBrowser function in lib/helpers.js explicitly disables the browser's security sandbox using the --no-sandbox and --disable-setuid-sandbox flags, increasing the risk of a host system compromise if a malicious website is visited.

playwright-skill