diagramming-domain-models
Pass
Audited by Gen Agent Trust Hub on Sep 27, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTIONCOMMAND_EXECUTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill processes external source code to generate model descriptions, creating a surface for indirect prompt injection.\n
- Ingestion points: The tool reads code files from domain-related paths (e.g.,
src/domain/) to extract architectural patterns and comments.\n - Boundary markers: While it strictly limits its read scope to domain layers, it does not employ specific delimiters for untrusted content within code comments.\n
- Capability inventory: The skill uses
Writeaccess to create diagrams andBashto run its internal validation scripts.\n - Sanitization: The tool escapes HTML for the final report. Although it filters model identifiers through a 'purity gate', it allows arbitrary text in
descriptionfields, which is a common trade-off for analysis utility.\n- [COMMAND_EXECUTION]: The tool executes internal scripts to perform deterministic model validation.\n - Evidence:
lib/render.mjsinvokes the companionlib/validate.mjsscript using Node.js subprocesses (spawnSync).\n - Context: This execution is scoped to the skill's own logic and uses absolute pathing to ensure only the intended validation script is run.
Audit Metadata