imessage-extraction

Fail

Audited by Gen Agent Trust Hub on Oct 10, 2026

Risk Level: HIGHPRIVILEGE_ESCALATIONDATA_EXFILTRATIONCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONDYNAMIC_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
  • [PRIVILEGE_ESCALATION]: The skill instructs the user to grant 'Full Disk Access' to their terminal application.
  • Evidence: Found in SKILL.md under the 'Full Disk Access required' section.
  • Impact: This configuration bypasses macOS Transparency, Consent, and Control (TCC) protections, allowing the agent to access sensitive directories that are normally protected by the OS.
  • [DATA_EXFILTRATION]: The skill is designed to access and extract data from highly sensitive local databases containing private user communications and contact information.
  • Evidence: SKILL.md and scripts/create_clean_db.py access ~/Library/Messages/chat.db (iMessage history) and ~/Library/Application Support/AddressBook/ (Contact details).
  • Detail: These files contain comprehensive histories of private messages, names, phone numbers, and email addresses.
  • [COMMAND_EXECUTION]: The skill makes extensive use of shell commands and database utilities to process user data.
  • Evidence: Documentation and scripts reference the execution of sqlite3 for database manipulation and the use of the yap command-line tool for audio transcription.
  • [INDIRECT_PROMPT_INJECTION]: The skill creates a vulnerability surface by processing untrusted data (messages sent by third parties) while possessing command execution capabilities.
  • Ingestion points: Reads from the message table in chat.db (file path mentioned in SKILL.md).
  • Boundary markers: None identified in instructions or scripts.
  • Capability inventory: Shell command execution (sqlite3, yap) and filesystem write access (scripts/create_clean_db.py).
  • Sanitization: While a clean_message function exists in scripts/create_clean_db.py, it focuses on removing binary artifacts and formatting rather than filtering for malicious natural language instructions.
  • [DYNAMIC_EXECUTION]: The skill performs dynamic SQL execution and runtime decoding of complex binary formats.
  • Evidence: scripts/create_clean_db.py utilizes target_cursor.executescript() to initialize database schemas and NSKeyedUnarchiver to process binary attributedBody blobs.
  • [EXTERNAL_DOWNLOADS]: The skill recommends and utilizes external software for specific processing tasks.
  • Evidence: Instructs the installation of the pyobjc-framework-Cocoa library via pip and references the yap transcription tool provided through the Homebrew package manager.
Recommendations
  • AI detected serious security threats
Audit Metadata
Risk Level
HIGH
Analyzed
Oct 10, 2026, 11:52 PM