verbatra-studio-agent-tools
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill handles untrusted translation strings and glossary terms from locale files, which presents a surface for indirect prompt injection attacks. Ingestion points: Tools such as verbatra_locale_values, verbatra_key_value, and verbatra_glossary_get ingest content from project files. Boundary markers: The skill includes a 'Non-negotiable rule' (Rule 4) that explicitly warns the agent to treat translatable strings as untrusted data and ignore any embedded instructions. Capability inventory: The skill can modify project files via verbatra_translation_editEntry and trigger paid API calls through verbatra_translation_translatePending. Sanitization: Integrity gates are applied to placeholders and ICU formats to prevent the writing of malformed or malicious values.
- [SAFE]: The skill strictly follows security best practices by requiring all provider API keys to be managed through system environment variables, prohibiting hardcoded secrets in files or command outputs. It also features automatic redaction of API-key-like patterns in glossary data to prevent accidental credential leakage.
Audit Metadata