The Agent Skills Directory

Command Execution: The skill interacts with the local environment by executing system commands like leaks and xcrun to manage simulators and analyze memory graphs. The implementation in scripts/summarize_memgraph_leaks.py uses subprocess.run to invoke these tools. Security is maintained by passing arguments as a list (avoiding shell execution) and using regular expressions to ensure that memory addresses extracted from the data are valid hexadecimal strings before they are used in further commands.
Indirect Prompt Injection Surface: The skill summarizes output from the leaks tool, which is generated from the memory contents of a running application. This provides a surface where data from an untrusted application is parsed and presented to the agent.
Ingestion points: The Python script scripts/summarize_memgraph_leaks.py ingests text output from the system leaks command.
Boundary markers: The rendered Markdown summary does not use specific delimiters or warnings to isolate the parsed leak data from the agent's instructions.
Capability inventory: The skill utilizes subprocess.run to call diagnostic utilities and performs local file read/write operations to store memory graphs and leak reports.
Sanitization: The script applies strict regular expression filtering to memory addresses (e.g., 0x[0-9a-fA-F]+) to ensure that only expected data types are processed from the tool output.

ios-memgraph-leaks