The Agent Skills Directory

[COMMAND_EXECUTION]: The skill utilizes browser-devtools-cli and node-devtools-cli to execute diagnostic and control commands against browser instances and Node.js processes.
[INDIRECT_PROMPT_INJECTION]: The skill ingests untrusted data from multiple external surfaces which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: Browser console messages (get-console-messages), network request/response metadata (get-http-requests), HTML content (get-as-html), and Node.js application logs (get-logs).
Boundary markers: The instructions do not define delimiters or warnings to ignore instructions embedded within the data retrieved from the target applications.
Capability inventory: The agent can perform browser navigation, simulate user interactions (clicks), take screenshots, and modify runtime execution via tracepoints and logpoints.
Sanitization: No evidence of sanitization, validation, or escaping of the retrieved external content before it is processed or presented in the agent context.
[DYNAMIC_EXECUTION]: The skill implements features that evaluate dynamic JavaScript strings within the target application context, such as watch expressions (--expression) and logpoints (--log-expression). While these are standard debugging functions, they represent the execution of arbitrary code logic supplied by the agent or potentially influenced by indirect injection.

debugging