alterlab-genai-voice-cloner
Pass
Audited by Gen Agent Trust Hub on Sep 22, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill instructions define an operational flow that ingests untrusted data, creating a vulnerability surface where malicious instructions in external content could influence the agent's behavior.
- Ingestion points: The agent is directed to "search the web" for current platform requirements and "read existing project files" including voice briefs, scripts, and documentation (SKILL.md).
- Boundary markers: There are no explicit delimiters or instructions to ignore commands embedded within the ingested files or search results.
- Capability inventory: The agent possesses the capability to search the web, read local files, and write structured markdown files to the project directory.
- Sanitization: The skill lacks any instructions for sanitizing or validating external content before it is interpolated into the agent's context or written to output files.
Audit Metadata