The Agent Skills Directory

[COMMAND_EXECUTION]: The skill uses rockcli sandbox exec to perform various administrative tasks within the sandbox, such as listing files, reading logs, and inspecting Docker containers. These commands are executed with shell privileges in the target environment.\n- [DYNAMIC_EXECUTION]: The skill utilizes python3 -c to execute Python code provided as string literals. This pattern is used for parsing JSON outputs and formatting reports. Additionally, the skill includes a complete Python script template for batch scanning Harbor jobs.\n- [OBFUSCATION]: The skill uses chr(39) within Python one-liners to represent single quotes. This technique is used to prevent shell injection or quoting issues when passing code to python3 -c, but it also serves to obscure the string literals being used.\n- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and process external data from log files and job results which could be influenced by external actors.\n
Ingestion points: Accesses result.json, trial.log, exception.txt, and other logs via cat commands.\n
Boundary markers: No specific delimiters or safety instructions are used when reading these files.\n
Capability inventory: Possesses significant capabilities including file system access and command execution via rockcli.\n
Sanitization: There is no evidence of sanitization or filtering applied to the content of the logs before they are processed by the agent.

rock-agent-debug