The Agent Skills Directory

[DATA_EXFILTRATION]: The script scripts/cleanup-report.sh performs a system-wide scan of running processes and listening network ports. This data collection includes process IDs, command-line arguments, and memory usage metrics, which are sensitive environment metadata.\n- [COMMAND_EXECUTION]: The skill's primary purpose is to facilitate the termination of processes by generating kill commands. While execution is intended to be user-initiated, the automated generation of these commands based on calculated 'waste scores' poses a risk of disrupting critical processes if misidentified.\n- [PROMPT_INJECTION]: The skill possesses an indirect prompt injection surface because it processes command-line strings from the system process list, which can be influenced by any running application. A malicious process could use a crafted name to deliver instructions to the agent during the reporting phase.\n
Ingestion points: System process command lines retrieved in scripts/cleanup-report.sh via the ps command.\n
Boundary markers: The collected data is structured into a JSON format before being presented to the agent.\n
Capability inventory: The agent utilizing this skill has access to the Bash tool and is tasked with recommending process termination.\n
Sanitization: The script implements character escaping for JSON validity but does not sanitize the process names for potential natural language instructions.\n- [EXTERNAL_DOWNLOADS]: The benchmark metadata configuration references a resource on the author's official GitHub repository (github.com/b-open-io). This is documented as a vendor-managed resource used for development and evaluation purposes.

process-cleanup