big-arrow

Warn

Audited by Gen Agent Trust Hub on Nov 10, 2026

Risk Level: MEDIUMPERSISTENCEDATA_EXFILTRATIONCOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [PERSISTENCE]: The skill provides instructions for setting up a persistent UserPromptSubmit hook within the AI agent's configuration file (~/.claude/settings.json). This hook ensures the bigarrow stop --hook command executes automatically upon every prompt submission. While used for UI cleanup, modifying agent configuration to establish persistent execution hooks is a known persistence mechanism.
  • [DATA_EXFILTRATION]: The bigarrow tool is designed to retrieve sensitive data from local credential managers, including the macOS Keychain (keychain:), 1Password (op:), and the pass utility, for display in a local GUI. This involves high-privilege read access to the user's secret stores.
  • [COMMAND_EXECUTION]: The skill documentation describes the use of several shell commands, including bigarrow, open (for system preferences), and brew (for software updates). There is a discrepancy between the commands shown in the documentation and the restrictive allowed-tools list in the YAML frontmatter which only specifies the bigarrow binary.
  • [EXTERNAL_DOWNLOADS]: The skill recommends installing and upgrading its required CLI tool via a vendor-owned Homebrew tap (franzenzenhofer/tap/bigarrow). These downloads originate from the skill author's own repository infrastructure.
  • [INDIRECT_PROMPT_INJECTION]: The skill's UI elements interpolate data using a custom template syntax ({{value}}). If the agent processes untrusted external data and passes it to the --text or --field arguments, it could be used to present deceptive copy-paste buttons or misleading instructions to the human user.
  • Ingestion points: Arguments for --text, --field, and JSON card definitions in SKILL.md and reference.md.
  • Boundary markers: Braces ({{ }}) are used for template parsing but do not serve as security boundaries or isolation markers.
  • Capability inventory: The bigarrow CLI can read local secrets, inspect UI elements, and modify terminal focus through its various subcommands.
  • Sanitization: The tool performs syntax validation for balanced braces but does not sanitize the content of the interpolated strings.
Audit Metadata
Risk Level
MEDIUM
Analyzed
Nov 10, 2026, 02:56 AM