The Agent Skills Directory

[COMMAND_EXECUTION]: The script check.sh executes several macOS system utilities, including ps, sysctl, vm_stat, df, and footprint. These tools are used to gather detailed information about system memory, disk usage, and running processes.\n- [DATA_EXFILTRATION]: The process monitoring logic specifically captures full command-line arguments (ps axo ... args=). This is a potential security risk as command-line arguments often inadvertently contain sensitive information such as API keys, authentication tokens, or database credentials. This data is output in a structured JSON format intended for consumption by an AI agent and potential delivery to external notification channels via a messaging tool.\n- [INDIRECT_PROMPT_INJECTION]: The skill represents an indirect prompt injection surface because it feeds untrusted data (process names and their command arguments) directly into an AI agent's context for report generation.\n
Ingestion points: System process names and command-line arguments captured in check.sh.\n
Boundary markers: None. The agent is provided a JSON blob and a report template in openclaw-cron.json without instructions to treat the data as untrusted or to ignore embedded instructions.\n
Capability inventory: The agent is authorized to run the check.sh script and use a message tool to transmit the resulting report.\n
Sanitization: None. The script enriches process names but does not filter or sanitize the command-line strings for potentially malicious content or hidden instructions.

system-watchdog