The Agent Skills Directory

[DYNAMIC_EXECUTION]: The skill provides a JavaScript template to be saved as a local file (e.g., /tmp/cdp-debug.js) and executed via Node.js. This script utilizes the chrome-remote-interface to interact with the Debugger and Runtime domains of a target process, including the evaluation of arbitrary expressions via Debugger.evaluateOnCallFrame and manual property inspection through Runtime.getProperties.
[COMMAND_EXECUTION]: The instructions involve using shell commands to manage processes, such as kill -SIGUSR1 to enable the inspector on a running PID and node inspect to attach the debugger CLI. It also suggests using pgrep to find process IDs.
[EXTERNAL_DOWNLOADS]: The skill directs the installation of the chrome-remote-interface package from the npm registry to enable programmatic debugging.
[INDIRECT_PROMPT_INJECTION]: The skill creates an attack surface by reading and evaluating state from external processes. If a process being debugged contains malicious data or code, it could potentially be used to influence the agent's context or behavior.
Ingestion points: Data and properties read from the target process via CDP (Runtime.getProperties, Debugger.evaluateOnCallFrame) which are then logged to the console.
Boundary markers: None present in the instructions or the provided script template.
Capability inventory: The skill has the ability to write to the local file system (fs.writeFileSync), execute shell commands, and perform arbitrary JS evaluation in a remote process.
Sanitization: No sanitization or validation of the data retrieved from the debugged process is described or implemented in the provided template.

node-inspect-debugger