spoken-response-controls

Pass

Audited by Gen Agent Trust Hub on Oct 6, 2026

Risk Level: SAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTIONINDIRECT_PROMPT_INJECTION
Full Analysis
  • [EXTERNAL_DOWNLOADS]: The skill instructs the user to install the agent-voice package using uv tool install agent-voice. This package is a dependency provided by the skill author and hosted on public repositories.
  • [COMMAND_EXECUTION]: The skill uses the agent-voice CLI to process text. It provides instructions to the agent to run shell commands like agent-voice speak "$RESPONSE_AS_TEXT". While the instructions include quoting the variable, this pattern of interpolating content that can be influenced by external data into a shell command presents a potential command injection risk if the shell environment allows subshell execution within double quotes.
  • [INDIRECT_PROMPT_INJECTION]: The skill processes assistant responses to generate audio, creating an attack surface for indirect prompt injection.
  • Ingestion points: The assistant's current or previous response text (defined in SKILL.md).
  • Boundary markers: None identified in the instructions.
  • Capability inventory: Shell command execution via agent-voice and file system writes to temporary directories (defined in SKILL.md).
  • Sanitization: The instructions suggest basic formatting cleanup (e.g., converting literal newlines to spacing) but do not include rigorous escaping or validation of the text before it is passed to the command line.
Audit Metadata
Risk Level
SAFE
Analyzed
Oct 6, 2026, 11:03 AM