amq-remote

Pass

Audited by Gen Agent Trust Hub on Sep 28, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTIONPERSISTENCE
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the user to install tools from avivsinai/tap/amq via Homebrew. These are vendor-owned resources belonging to the skill's author.
  • [COMMAND_EXECUTION]: The skill utilizes several CLI tools (amq-remote, amq-acp, amq) and executes shell commands to configure the environment and handle messages.
  • [PERSISTENCE]: The skill provides instructions to run amq-remote claude install-stop-hook, which modifies the ~/.claude/settings.json configuration file to install a persistent stop hook for the agent environment. This is a documented step for integration.
  • [INDIRECT_PROMPT_INJECTION]: The skill's primary function is to act as a receiver for external commands sent via the Buzz messaging system, which introduces an injection risk.
  • Ingestion points: Incoming messages from the user buzz with the subject "Buzz DM" are processed from the agent's AMQ inbox.
  • Boundary markers: The instructions do not specify any delimiters or safety prompts to isolate the external message content from the agent's internal logic.
  • Capability inventory: The agent is explicitly told to "Do the work" requested in the message, which grants the external sender access to the agent's full range of tools and command execution capabilities.
  • Sanitization: No sanitization, validation, or filtering of the incoming message content is described.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 28, 2026, 07:54 AM