fable-domain
Pass
Audited by Gen Agent Trust Hub on Oct 10, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes untrusted data from web search and user requirements to generate new instructions, adapters, and evaluation scenarios.
- Ingestion points: Untrusted data enters the agent context via the
user_requirementsinput and bounded web research results extracted from authoritative sources (Stage 2 in SKILL.md). - Boundary markers: The instructions do not define specific delimiters or "ignore embedded instructions" warnings when interpolating researched data into the generated domain adapters or workflows.
- Capability inventory: The skill has permissions to modify the local environment (
mutatesWorkspace: true) and writes generated files toreferences/domains/,eval/scenarios/, and documentation registry paths. - Sanitization: The skill implements "Red-Line Refusal" and "Scope Gate" logic to prevent processing high-harm domains such as medical diagnosis, legal counsel, or financial advice.
Audit Metadata