The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The skill configuration instructs the agent to download and run the chrome-devtools-mcp package using npx -y. While targeting a recognized package from the Chrome DevTools team, this pattern involves executing code fetched from a remote registry.
[PROMPT_INJECTION]: The skill is designed to navigate to and process content from arbitrary web pages, which is a primary vector for indirect prompt injection.
Ingestion points: Untrusted data enters the context through navigate_page and element snapshots provided by take_snapshot in SKILL.md.
Boundary markers: The instructions do not define delimiters or specific warnings to ignore instructions embedded within the page content.
Capability inventory: The skill provides extensive capabilities including evaluate_script, click, fill_form, and upload_file as described in tool-reference.md.
Sanitization: There is no evidence of sanitization or filtering of the content retrieved from web pages before it is processed by the agent.
[COMMAND_EXECUTION]: The evaluate_script tool allows for the execution of arbitrary JavaScript functions within the context of the loaded web page. This provides a high degree of control over the browser environment.
[DATA_EXFILTRATION]: The skill allows for the inspection of sensitive browser data, including network logs via list_network_requests, console messages, and page content. The documentation correctly identifies privacy risks associated with persistent browser profiles and recommends the use of clean or isolated profiles to mitigate accidental data exposure.

chrome-devtools