gpt-live-demo
Pass
Audited by Gen Agent Trust Hub on Sep 14, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes user-provided voice transcriptions and interpolates them directly into prompts sent to a backend agent. This lacks proper boundary markers or sanitization, potentially allowing malicious user input to influence the agent's behavior.
- Ingestion points:
scripts/server.mjs(thetextfield in the/api/agentendpoint). - Boundary markers: Absent. User input is concatenated into the system prompt string.
- Capability inventory: The skill writes to local JSON files (
data/tasks.json) and performs network requests to a local agent (localhost:8000). - Sanitization: Absent. Input is trimmed but not sanitized or escaped.
- [COMMAND_EXECUTION]: The skill facilitates the execution of tasks by delegating to a local agent via an HTTP SSE stream at
localhost:8000. While this is an intended feature of the Starchild platform, it represents a capability that could be abused if the backend agent is not properly secured.
Audit Metadata