prose-reviser

Pass

Audited by Gen Agent Trust Hub on Nov 13, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [INDIRECT_PROMPT_INJECTION]: The skill processes external, user-supplied content such as "source text" and "style guides," which introduces a potential surface for indirect prompt injection where instructions embedded in the data could attempt to influence the agent's behavior.
  • Ingestion points: The agent accepts and processes external data via the "source text," "supplied style or voice guide," and other user-provided inputs as defined in the Inputs section.
  • Boundary markers: The instructions do not define specific delimiters (e.g., XML tags or triple quotes) or provide explicit commands for the agent to ignore instructions embedded within the source prose.
  • Capability inventory: The skill possesses the capability to write or edit files if the user provides a target path, as noted in the Output section.
  • Sanitization: The skill implements a form of data protection by identifying "protected spans" (code, commands, identifiers, links, and structured data) that must remain intact, though this does not explicitly sanitize against logical instructions embedded in the prose.
Audit Metadata
Risk Level
SAFE
Analyzed
Nov 13, 2026, 09:38 PM