synthesis-mac-sync

Warn

Audited by Gen Agent Trust Hub on May 12, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONREMOTE_CODE_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill instructs the agent to dynamically generate and execute complex, multi-line bash scripts. These scripts perform operations such as directory creation, recursive file discovery, and conditional file synchronization across the local file system.
  • [REMOTE_CODE_EXECUTION]: The 'One-time actions' feature defines a system where arbitrary shell commands are retrieved from a remote manifest file (README.md in iCloud) and executed on the local machine. This allows for the execution of instructions sourced outside the skill's static definition.
  • [DATA_EXFILTRATION]: The core functionality of the skill involves reading sensitive configuration files—such as .gitconfig, .zshrc, and application-specific key files containing secrets—and uploading them to iCloud Drive for cross-machine synchronization.
  • [PROMPT_INJECTION]: The skill demonstrates a surface for indirect prompt injection by ingesting and acting upon instructions within external data files (git-repos.yaml and the iCloud README.md).
  • Ingestion points: Config files and manifests located in the iCloud sync directory.
  • Boundary markers: No explicit delimiters or instructions are provided to the agent to ignore embedded instructions in the synced data.
  • Capability inventory: The skill utilizes bash tool execution, full file system access within the user home directory, and Git network operations.
  • Sanitization: The skill does not implement validation or sanitization for the commands or data read from the external manifest files.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 12, 2026, 09:44 AM
Security Audit — agent-trust-hub — synthesis-mac-sync