voice-builder
Pass
Audited by Gen Agent Trust Hub on Aug 3, 2026
Risk Level: SAFEPROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill utilizes imperative instructions in the "CRITICAL: Auto-start on load" section to override the agent's default interaction flow. It strictly prohibits the agent from providing summaries, explanations, or confirmation prompts, forcing an immediate transition to the interview process in Step 1.
- [PROMPT_INJECTION]: The skill exposes an indirect prompt injection surface when processing writing samples in Step 3 and Step 4.
- Ingestion points: Writing samples provided by the user in Step 3 or loaded from
references/sample-content.md. - Boundary markers: None identified in the instructions to separate sample data from instructions.
- Capability inventory: The agent can write files to the project root (
about-me.md,voice.md) and use theAskUserQuestiontool to interact with users. - Sanitization: No sanitization or filtering is performed on the input samples before analysis.
Audit Metadata