voice-preserving-rewriter

Pass

Audited by Gen Agent Trust Hub on Sep 9, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill ingest untrusted data in the form of user requests and draft text, which creates a potential surface for indirect prompt injection.
  • Ingestion points: Untrusted data enters the agent context through the user request and input text identified in Step 1 of the Workflow in SKILL.md.
  • Boundary markers: The skill instructions protect specific content types (e.g., quoted material, code blocks) but do not define global delimiters or specific instructions to ignore embedded commands in the source text.
  • Capability inventory: The skill is scoped to text transformation; the provided scripts and instructions do not include subprocess calls, direct file writes, or network operations, limiting the impact of any potential injection.
  • Sanitization: While the skill contains an Anti-Pattern warning against treating external content as trusted instructions, it does not detail specific input sanitization or filtering logic for the text being rewritten.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 9, 2026, 03:52 AM
Security Audit — agent-trust-hub — voice-preserving-rewriter