The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/paired-probe.sh accepts shell commands via the --present, --absent, and --measure arguments. These commands are executed directly by the script, allowing the AI to run arbitrary shell operations under the guise of verification.
Evidence: scripts/paired-probe.sh lines 79, 107, 108: out=$(eval "$1" 2>&1).
[DYNAMIC_EXECUTION]: The script uses the eval function in Bash to execute strings provided at the command line. This pattern is often a source of injection vulnerabilities if inputs are sourced from untrusted data, though in this case, the input typically comes from the agent's own instructions.
Evidence: The run() function in scripts/paired-probe.sh (line 79) performs eval "$1" on whatever string is passed to it.

paired-probe