The Agent Skills Directory

[COMMAND_EXECUTION]: The script scripts/perf_profile.py executes arbitrary code from local files provided as command-line arguments.
Evidence: The script uses exec() to run compiled bytecode in run_cpu_profile, run_memory_profile, run_combined_profile, and run_line_profile functions.
[REMOTE_CODE_EXECUTION]: The tool enables the execution of externally provided Python source code. If an agent is instructed to profile a script from an untrusted source, the script will execute with the agent's current process privileges.
Evidence: In _prepare_exec and run_line_profile, the skill reads a file's content and converts it into executable code via the compile() function.
[INDIRECT_PROMPT_INJECTION]: The skill exhibits an attack surface for indirect prompt injection because it processes and executes untrusted data (Python scripts).
Ingestion points: The script argument in scripts/perf_profile.py specifies a file path to be read and executed.
Boundary markers: None. The script source is read and executed in its entirety.
Capability inventory: The skill has access to the file system (via pathlib and open) and the ability to execute code (exec, compile).
Sanitization: No validation or sandboxing is performed on the code being profiled.

py-perf-analyzer