The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the agent to execute various system profiling utilities including strace, perf, nvidia-smi, iostat, vmstat, and py-spy. It also requires executing the user-provided target specified in the $ARGUMENTS variable.
[REMOTE_CODE_EXECUTION]: The instructions explicitly direct the agent to "write and insert profiling code into the target" and "run the chosen methods." This involves the dynamic generation and execution of arbitrary code (instrumentation) which can modify existing application logic.
[DATA_EXFILTRATION]: While no external network calls are hardcoded, the skill is designed to read and process sensitive system information, including process IDs (PIDs), hardware topology, and source code, in order to generate reports.
[PROMPT_INJECTION]: The skill processes untrusted input through the $ARGUMENTS variable and the content of the files it instruments.
Ingestion points: User-provided $ARGUMENTS and local source code files designated for profiling.
Boundary markers: None specified in the instructions to prevent the agent from following instructions embedded in the target code.
Capability inventory: Full subprocess execution, file modification, and shell command access across the system profiling toolset.
Sanitization: No validation or sanitization is performed on the $ARGUMENTS or the target code before execution or instrumentation.

system-profile