The Agent Skills Directory

[COMMAND_EXECUTION] (MEDIUM): The skill extracts an arbitrary command string from user-provided arguments for execution. Evidence: The JavaScript parsing block in SKILL.md identifies any non-numeric string as the command variable. Risk: If the underlying agent executes this string in a shell environment, it is susceptible to command injection attacks using shell operators like semicolons or pipes.
[PROMPT_INJECTION] (LOW): The skill exhibits an indirect prompt injection surface. 1. Ingestion points: The command argument parsed from $ARGUMENTS. 2. Boundary markers: No delimiters or ignore-instructions are applied to the input command, though PERF_METRICS_START/END markers are used for output. 3. Capability inventory: Execution of external system commands. 4. Sanitization: The skill performs no validation, escaping, or allowlisting of the command string before processing.

perf-benchmarker