The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill is designed to scan untrusted local repositories using shell commands and report the findings back to the agent context.
Ingestion points: Local source code files (.swift, .m, .h) and configuration files (.plist, .xib, .storyboard) provided by the user for auditing.
Boundary markers: The instructions lack explicit delimitation or warnings to prevent the agent from following instructions hidden within the analyzed source code (e.g., malicious comments designed to influence the audit report).
Capability inventory: The skill utilizes grep, PlistBuddy, xcodebuild, ls, and curl to perform its tasks.
Sanitization: While the skill provides specific regex patterns for its scans, the agent's interpretation of the scan results remains a surface for indirect injection.
[COMMAND_EXECUTION]: The skill provides a large suite of shell commands for the agent to execute on the host system to perform environmental checks and code analysis.
Evidence: references/audit-checklist.md contains complex grep and PlistBuddy chains. references/testing-and-sources.md includes xcodebuild, xcrun, and simctl commands for toolchain verification.
[EXTERNAL_DOWNLOADS]: The skill fetches information from external web services to verify the freshness of its simulated API data and toolchain status.
Evidence: references/testing-and-sources.md contains curl commands targeting developer.apple.com, apple.com, and xcodereleases.com.
Context: These operations target trusted or well-known developer services and are used for data verification rather than direct code execution.
[DYNAMIC_EXECUTION]: The skill defines and executes shell functions at runtime to facilitate its auditing process.
Evidence: references/audit-checklist.md instructs the agent to declare and use the G() and GM() bash functions to streamline recursive code scanning.

iphone-duo