sap-content-generator

Pass

Audited by Gen Agent Trust Hub on Mar 6, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: An indirect prompt injection surface exists as the skill processes external data (e.g., protocol_entities.json) and interpolates it directly into LLM prompts without explicit sanitization or strict boundary markers.
  • Ingestion points: scripts/generate_by_chapter.py (loading template structure) and all prompt templates in the prompts/ directory.
  • Boundary markers: Absent; the prompts use simple variable interpolation (e.g., {endpoints}, {study_design}) which does not distinguish between system instructions and potentially adversarial data content.
  • Capability inventory: The skill is capable of reading and writing local files (Markdown sections and JSON reports) using the provided Python scripts.
  • Sanitization: scripts/generate_by_chapter.py contains basic regex-based filename sanitization (_sanitize_id), but there is no validation or escaping of the actual content processed by the LLM.
Audit Metadata
Risk Level
SAFE
Analyzed
Mar 6, 2026, 04:20 PM