The Agent Skills Directory

[DATA_EXFILTRATION]: The skill instructs the agent to inspect sensitive system directories that store privacy and authorization records for the macOS Transparency, Consent, and Control (TCC) framework.
Evidence: Step 3 in SKILL.md directs listing the contents of ~/Library/Application Support/com.apple.TCC/ and /Library/Application Support/com.apple.TCC/.
Context: The skill contains explicit instructions to the agent that TCC and Keychain data are private and must not be exported, referring to exfiltration as a forbidden "red line" behavior. The access is strictly for read-only analysis within the reverse engineering workflow.
[INDIRECT_PROMPT_INJECTION]: The skill analyzes external, untrusted application bundles and binaries, which presents an attack surface for indirect prompt injection if the processed data contains malicious instructions.
Ingestion points: App bundles (Sample.app), Mach-O binaries, and Info.plist files processed by system tools.
Boundary markers: None identified in the prompt interpolation steps.
Capability inventory: Shell command execution (codesign, otool, spctl, defaults), Mach-O parsing, and dynamic debugging (lldb).
Sanitization: No specific validation or sanitization of file content is mandated before analysis with command-line utilities.
[EXTERNAL_DOWNLOADS]: The skill recommends installing external security and development tools.
Evidence: Mentions installation of the LLVM toolchain via Homebrew (brew install llvm) and Apple's Xcode Command Line Tools (xcode-select --install).
Context: These references target well-known and trusted package managers and official vendor services, which is consistent with standard development and security research practices.

re-macos