The Agent Skills Directory

[COMMAND_EXECUTION]: The skill describes how to use Apple's profiling utility xctrace along with supporting tools like man, grep, and dwarfdump for UUID verification. These are legitimate system commands for performance optimization.
[INDIRECT_PROMPT_INJECTION]: The skill establishes a surface for processing external trace files. Ingestion points: Apple Trace files (.trace and .atrc) referenced in references/xctrace-overview.md. Boundary markers: None present. Capability inventory: Execution of xctrace subcommands (toc, list tracks, show, triage). Sanitization: Not applicable as the agent is expected to interpret the command-line output from a system utility.
[PROMPT_INJECTION]: The instructions include statements intended to supersede the model's existing knowledge with the provided technical data. This is used here in a benign manner to ensure the model uses correct subcommand flags and data interpretation logic.
[METADATA_POISONING]: The skill claims to be published by Apple in its description and introductory text, despite being attributed to a different author. While this is a discrepancy in metadata, it does not impact the security of the profiling tasks described.

xctrace