translate-content

Warn

Audited by Gen Agent Trust Hub on May 20, 2026

Risk Level: MEDIUMCOMMAND_EXECUTIONDATA_EXFILTRATIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill executes local Python scripts (scripts/language-router.py and scripts/brand-voice-scorer.py) using shell commands that interpolate user-provided text and file paths. This pattern is vulnerable to command injection if the input contains shell metacharacters such as semicolons or backticks.
  • [DATA_EXFILTRATION]: The skill accesses brand profiles and guidelines stored in ~/.claude-marketing/ and transmits this information to well-known translation services (DeepL, Google Cloud, Sarvam AI, Lara Translate) to ensure brand voice consistency.
  • [PROMPT_INJECTION]: The skill is susceptible to indirect prompt injection because it ingests untrusted content for translation without using boundary markers or sanitization.
  • Ingestion points: User-supplied text or file content via the 'Content to translate' input.
  • Boundary markers: None present in the process description.
  • Capability inventory: Execution of subprocesses (Python scripts), reading from the local filesystem (~/.claude-marketing/), and making network requests via multiple MCP servers.
  • Sanitization: No sanitization or escaping of the input text is performed before it is processed by the scripts or translation services.
Audit Metadata
Risk Level
MEDIUM
Analyzed
May 20, 2026, 12:41 AM
Security Audit — agent-trust-hub — translate-content