The Agent Skills Directory

[SAFE]: No malicious patterns, obfuscation, base64-encoded payloads, or persistence mechanisms were detected in the skill instructions or scripts.
[SAFE]: The skill does not perform any network operations, ensuring that project configuration data is not exfiltrated to external servers.
[SAFE]: Command execution is limited to local auditing using standard tools like grep and a bundled Python script that processes project metadata using strict regular expressions.
[SAFE]: The skill uses a safety-oriented workflow where all proposed project changes are staged in a plan file (xcode-security-audit-plan.md) for explicit user review and approval before execution.
[SAFE]: The indirect prompt injection attack surface (processing untrusted project files) is mitigated by the human-in-the-loop approval step and the use of non-executable data extraction methods.
Ingestion points: Reads project configuration files such as project.pbxproj, .entitlements, and .xcconfig.
Boundary markers: Staging changes in a plan file for user approval creates a clear verification boundary.
Capability inventory: Employs MCP tools for targeted file reading and writing, with no access to sensitive system directories or credentials.
Sanitization: Uses pre-defined regex filters to extract specific build settings, avoiding the execution of untrusted file content.

audit-xcode-security-settings