The Agent Skills Directory

[EXTERNAL_DOWNLOADS]: The skill instructs the user to use npx -y chrome-devtools-axi to download and execute code from an external repository (kunchenguid/chrome-devtools-axi) which is not a recognized trusted source or vendor repository.
[REMOTE_CODE_EXECUTION]: Executing the chrome-devtools-axi package via npx allows for arbitrary code execution from an externally managed package on the host machine. The skill also mentions fetching chrome-devtools-mcp@latest from the ChromeDevTools organization.
[DYNAMIC_EXECUTION]: The run subcommand is described as dynamically importing and executing temporary scripts on the host machine using absolute native paths. This mechanism allows the execution of host-side JavaScript, presenting a risk if script content is influenced by untrusted data.
[INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process data from live browser environments, which is an untrusted source.
Ingestion points: The skill reads page DOM, console logs, and network traffic via CLI commands (e.g., console-get, network-get, snapshot).
Boundary markers: Instructions explicitly label browser output as untrusted task data and warn against executing host commands or disclosing secrets based on page content.
Capability inventory: The agent has access to shell command execution (AXI CLI), file system writes (screenshots, traces, heap files), and host script execution (AXI run).
Sanitization: The skill recommends redacting credentials from request/response bodies and treating page-derived values strictly as data rather than executable scripts.
[COMMAND_EXECUTION]: The workflow relies heavily on CLI commands and the manipulation of environment variables such as CHROME_DEVTOOLS_AXI_SESSION, CHROME_DEVTOOLS_AXI_USER_DATA_DIR, and CHROME_DEVTOOLS_AXI_BROWSER_URL to control browser sessions and connection endpoints.

chrome-devtools-axi