call-when-afk
Pass
Audited by Gen Agent Trust Hub on Aug 22, 2026
Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
- [COMMAND_EXECUTION]: The skill invokes the
screenless callCLI tool to ring the user and wait for a verbal response. This command is executed within the terminal environment to bridge the gap when the user is away from the keyboard. - [PROMPT_INJECTION]: The agent is instructed to process and obey transcripts generated from voice calls, creating a vulnerability to indirect prompt injection.
- Ingestion points: The JSON
transcriptoutput from thescreenless callcommand inSKILL.md. - Boundary markers: Absent. There are no instructions to wrap the transcript in delimiters or to treat the voice input as potentially untrusted data.
- Capability inventory: The agent acts on the information provided in the transcript to make decisions and continue working on the task, which could involve any tools available to the agent.
- Sanitization: Absent. The skill does not provide mechanisms to sanitize or validate the verbal input before the agent interprets it as an instruction.
Audit Metadata