omnifocus

Warn

Audited by Gen Agent Trust Hub on Feb 27, 2026

Risk Level: MEDIUMDATA_EXFILTRATIONCOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [DATA_EXFILTRATION]: The skill accesses the local OmniFocus SQLite database which contains private user information including tasks, notes, and project details.
  • Evidence: scripts/sqlite_reader.rb reads files from sensitive paths like ~/Library/Caches/com.omnigroup.OmniFocus/OmniFocusDatabase2 and application containers.
  • [COMMAND_EXECUTION]: The skill uses osascript to execute dynamically constructed AppleScript and JavaScript (JXA) strings to automate the OmniFocus application.
  • Evidence: scripts/applescript_client.rb and scripts/omni_automation.rb generate script strings incorporating user-provided data and execute them via system shell backticks.
  • [PROMPT_INJECTION]: The skill is exposed to indirect prompt injection as it ingests and processes untrusted content from the user's OmniFocus database.
  • Ingestion points: Task names, notes, and project descriptions retrieved from the local database in scripts/omnifocus_manager.rb.
  • Boundary markers: While the output is formatted as JSON, the individual text fields do not include delimiters or instructions to ignore embedded commands.
  • Capability inventory: The skill has the capability to modify the local database, create new tasks, and execute shell commands through the Ruby environment.
  • Sanitization: The code uses Shellwords.escape to prevent shell injection but does not filter for malicious natural language instructions within the retrieved data.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Feb 27, 2026, 03:38 PM