brand-voice-extractor
Pass
Audited by Gen Agent Trust Hub on Nov 16, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill is designed to ingest and analyze untrusted external data (3-5 content samples provided by the user) to identify voice traits. This process introduces a risk where malicious instructions hidden within those samples could be processed by the agent.
- Ingestion points: External content enters the agent's context through user-pasted text or through files read by the
analyze-voice.mjsscript via the--sampleargument. - Boundary markers: The instructions in
SKILL.mddo not prescribe the use of clear delimiters (such as XML tags or unique markers) to isolate the untrusted sample content from the agent's instructions. - Capability inventory: The skill possesses capabilities for file system operations (reading and writing voice profiles in the
voice/directory) and executing local Node.js scripts. - Sanitization: There is no evidence of specific sanitization or filtering of the content samples; the skill performs regex-based statistical analysis (counting word frequencies and punctuation) but otherwise processes the raw text.
- [COMMAND_EXECUTION]: The skill uses a local Node.js script (
scripts/analyze-voice.mjs) to assist in the extraction process. While the script itself is benign and performs only text analysis using standard libraries, the agent is instructed to execute this tool with arguments (such as file paths) that may be influenced by user-provided input.
Audit Metadata