The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides scripts (record_trace.py and analyze_trace.py) that invoke the system-level xctrace utility. These scripts are used to list devices, record application activity, and export trace data for performance analysis.\n
Evidence: scripts/record_trace.py and scripts/instruments_parser/xctrace.py utilize the subprocess module to execute commands using the xctrace binary with arguments derived from user and agent input.\n- [INDIRECT_PROMPT_INJECTION]: The skill analyzes external data from .trace files, including application logs and signposts. This data is then processed and displayed to the agent, creating a vulnerability where malicious content within a trace file could influence the agent's logic or subsequent actions.\n
Ingestion points: scripts/analyze_trace.py processes .trace files (provided as paths) by exporting their contents to XML and parsing them.\n
Boundary markers: There are no explicit delimiters or instructions to treat trace content as untrusted data, which may lead the agent to follow instructions embedded in logs.\n
Capability inventory: The agent can execute local scripts that interact with the system (e.g., launching apps via xctrace) and is primary intended to perform code modifications on the project.\n
Sanitization: The provided Python tools use secure argument handling for process execution, but there is no specific sanitization of the text data extracted from the traces before it is presented to the agent.

swiftui-expert-skill