fable-domain
Pass
Audited by Gen Agent Trust Hub on Jul 15, 2026
Risk Level: SAFECOMMAND_EXECUTIONEXTERNAL_DOWNLOADS
Full Analysis
- [COMMAND_EXECUTION]: The skill executes local repository scripts during its verification phase, specifically 'the repo's own check script' and 'fable-judge' in suite mode to perform smoke evaluations on generated artifacts.
- [EXTERNAL_DOWNLOADS]: The skill performs 'bounded web research' in Stage 2 to fetch domain-specific evidence, practitioner authorities, and documented failure modes. This data is used to ground the generation of the domain bundle.
- [PROMPT_INJECTION]: The skill exhibits a surface for indirect prompt injection by ingesting untrusted data from the web during the research phase.
- Ingestion points: Stage 2 'Research' fetches external data including regulations, platform policies, and failure modes.
- Boundary markers: The skill uses an internal 'GROUND-TRUTH.md' (within the trap fixture) which is explicitly restricted from being shared with agents under test to prevent leakage.
- Capability inventory: The skill has the capability to write files to the repository (adapters, fixtures, flowcharts) and execute local scripts (fable-judge, check scripts).
- Sanitization: Stage 1 includes a 'Red-lines' hard refusal check that prevents the skill from generating content for domains requiring professional licensure or involving high physical, legal, or financial harm.
Audit Metadata