phone-call
Pass
Audited by Gen Agent Trust Hub on Aug 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCREDENTIALS_UNSAFEEXTERNAL_DOWNLOADSCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill accepts user-provided text for the phone call purpose which is directly used to construct the system prompt for an automated phone agent. This creates a surface for indirect prompt injection if the objective source is untrusted, potentially allowing a malicious payload to redirect the agent to perform social engineering. Ingestion points: scripts/phone_call.sh and scripts/create_agent.py process user purpose strings. Boundary markers: The skill does not implement delimiters to separate user input from the agent safety logic. Capability inventory: Initiates outbound phone calls via api.fluents.ai. Sanitization: No input validation is performed on the objective strings.
- [CREDENTIALS_UNSAFE]: The scripts/diagnose.py utility prints the first 10 and last 4 characters of the FLUENTS_API_KEY to standard output, exposing partial secrets in terminal history and logs.
- [EXTERNAL_DOWNLOADS]: The skill communicates with api.fluents.ai to manage agents and initiate calls, which is a recognized vendor resource for this skill.
- [COMMAND_EXECUTION]: The scripts/phone_call.sh shell script executes internal Python scripts to manage the phone call lifecycle.
Audit Metadata