voice-builder
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEPROMPT_INJECTIONINDIRECT_PROMPT_INJECTION
Full Analysis
- [PROMPT_INJECTION]: The skill includes a 'CRITICAL: Auto-start on load' section that uses forceful language ('MUST', 'Do NOT', 'Do THIS') to override the agent's standard operating procedures. It explicitly commands the agent to suppress summaries, explanations, and user confirmations that typically occur when a skill is loaded or triggered.
- [INDIRECT_PROMPT_INJECTION]: The skill's primary function involves ingesting and analyzing untrusted writing samples provided by the user in Step 3. These samples are then used to generate project files (
about-me.mdandvoice.md). - Ingestion points: User-provided writing samples pasted into the chat or loaded from reference files.
- Boundary markers: The instructions lack explicit delimiters or warnings to treat the writing samples as data rather than instructions during the analysis phase (Step 4).
- Capability inventory: The skill has the ability to write files to the project root (
about-me.mdandvoice.md). - Sanitization: There is no mention of sanitizing or escaping the content of the writing samples before they are processed by the LLM for voice analysis.
Audit Metadata