s4h-ethics-council

Pass

Audited by Gen Agent Trust Hub on Jun 16, 2026

Risk Level: SAFEPROMPT_INJECTION
Full Analysis
  • [PROMPT_INJECTION]: The skill demonstrates an attack surface for indirect prompt injection through its ingestion of untrusted workspace data.
  • Ingestion points: The agent is instructed in SKILL.md to scan the workspace for files like CLAUDE.md and other documentation to ground the ethical framing.
  • Boundary markers: The skill employs triple-dash (---) delimiters within its subagent prompt templates to isolate the external question from the core instructions.
  • Capability inventory: The skill possesses the ability to create and write local files (HTML and Markdown).
  • Sanitization: While the agent is prompted to frame questions neutrally, no explicit escaping or content validation for the ingested data is defined.
Audit Metadata
Risk Level
SAFE
Analyzed
Jun 16, 2026, 11:43 PM
Security Audit — agent-trust-hub — s4h-ethics-council