translate

Pass

Audited by Gen Agent Trust Hub on Jul 12, 2026

Risk Level: SAFECOMMAND_EXECUTIONPROMPT_INJECTION
Full Analysis
  • [COMMAND_EXECUTION]: The skill utilizes the Bash tool to execute local Python scripts (split_md.py, merge_md.py) and perform directory cleanup (rm -rf _parts/). These operations are part of the intended document processing workflow and operate exclusively on paths relative to the source file or skill directory.
  • [PROMPT_INJECTION]: The skill handles user-provided Markdown and text files, which presents an indirect prompt injection surface where instructions embedded in source text could target the translation agent.
  • Ingestion points: Untrusted data enters the agent context when reading source files in Phase 1 and Phase 2.
  • Boundary markers: The skill uses a structured sub-agent prompt (references/subagent-prompt.md) and a temporary translation brief (_translation_brief.md) to clearly define the boundaries of the text to be processed.
  • Capability inventory: The skill has permissions to read/write files and execute local scripts, but lacks network access for external data exfiltration.
  • Sanitization: A robust verification phase (Phase 3) includes checklists for format compliance, line counts, and terminology consistency to detect and mitigate potential manipulation before merging results.
Audit Metadata
Risk Level
SAFE
Analyzed
Jul 12, 2026, 12:05 AM
Security Audit — agent-trust-hub — translate