The Agent Skills Directory

[REMOTE_CODE_EXECUTION]: The entry script bin.mjs uses npx -y to download and execute the agent-lynx package from the npm registry during execution. This facilitates runtime code delivery from an external source based on the version specified in the skill's package.json.- [DYNAMIC_EXECUTION]: The skill provides the evaluate command and documentation for the Runtime.evaluate CDP method, enabling arbitrary JavaScript execution within the target Lynx application's background or main-thread VMs.- [COMMAND_EXECUTION]: The skill uses the agent-lynx CLI for primary operations including client discovery, session management, and interaction with connected devices.- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data which could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: DevTool console logs (get-console), ReactLynx component trees (reactlynx tree), and UI metadata (UITree.getLynxUITree).
Boundary markers: The instructions do not define delimiters or specific 'ignore' directives for processed app data.
Capability inventory: Arbitrary JS execution (evaluate), shell command execution (agent-lynx), and host reflection calls (App.CallStaticVoidMethod).
Sanitization: No automated sanitization or validation of data retrieved from the target application is documented.- [PRIVILEGE_ESCALATION]: The App.CallStaticVoidMethod command provides a mechanism to invoke static void methods on the host side via reflection, which may bypass standard API boundaries if sensitive methods are accessible in the host app environment.

lynx-devtool