verbatra-cli
Pass
Audited by Gen Agent Trust Hub on Oct 7, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill identifies that locale files containing source and translated strings act as untrusted inputs that could attempt to influence agent behavior.
- Ingestion points: Locale files in various formats (JSON, YAML, XLIFF, PO, etc.) processed by the CLI.
- Boundary markers: The instructions explicitly state that locale content must be treated as data and never as instructions.
- Capability inventory: The skill has capabilities to write files (locale files, config files) and call external translation APIs (Anthropic, OpenAI, Gemini, etc.).
- Sanitization: The skill instructs the agent to ignore any text inside a locale file that reads like a command.
- [CREDENTIALS_UNSAFE]: The skill enforces secure credential management practices by forbidding the inclusion of secrets in code or command lines.
- Evidence: Rule 1 in SKILL.md mandates that API keys for translation providers (Anthropic, OpenAI, etc.) must only be read from environment variables and never written to files or output.
Audit Metadata