The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill facilitates the analysis of untrusted external data sources, including iOS crash reports (.ips, .crash files) and MetricKit diagnostic payloads (MXCrashDiagnostic, MXHangDiagnostic). This introduces a potential surface where an attacker could embed malicious natural language instructions within stack traces, thread names, or diagnostic metadata to manipulate the agent's behavior during triage.
Ingestion points: As described in references/crash-triage.md, the skill acquires crash reports from multiple sources including User Analytics data, MetricKit telemetry, and the Xcode Organizer.
Boundary markers: The instructions do not specify any delimiters or explicit 'ignore embedded instructions' warnings for the agent when it reads and processes these diagnostic files.
Capability inventory: The skill leverages powerful system-level developer tools and shell commands, such as xctrace, atos, dwarfdump, and mdfind, and instructs the agent to run the CrashSymbolicator.py script to process the ingested data.
Sanitization: There are no defined mechanisms in the skill to sanitize, escape, or validate the content of the crash logs or diagnostic payloads before they are interpreted by the agent.

ios-performance-engineering