spoken-response-desktop

Pass

Audited by Gen Agent Trust Hub on Sep 21, 2026

Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill uses shell interpolation to pass the assistant's response to the agent-voice tool via the $RESPONSE_AS_TEXT variable. This pattern can be exploited for command injection if the content contains shell metacharacters like backticks or subshell expansion markers, especially when enclosed in double quotes.\n- [EXTERNAL_DOWNLOADS]: The setup process requires installing the agent-voice package from a remote repository using uv tool install. This package is a resource associated with the skill author.\n- [INDIRECT_PROMPT_INJECTION]: The skill processes assistant-generated text through a shell-based capability, creating an injection surface.\n
  • Ingestion points: The RESPONSE_AS_TEXT variable is populated from the assistant's response in SKILL.md.\n
  • Boundary markers: No delimiters or safety instructions are provided to ensure the content does not interfere with shell command structure.\n
  • Capability inventory: The skill utilizes the agent-voice CLI for audio synthesis and file operations as seen in SKILL.md.\n
  • Sanitization: There is no evidence of escaping or validation of the text before it is executed within a shell environment.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 21, 2026, 09:13 AM
Security Audit — agent-trust-hub — spoken-response-desktop