feature-design-assistant

Pass

Audited by Gen Agent Trust Hub on Sep 15, 2026

Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
  • [SAFE]: The skill follows a legitimate interaction model, using structured JSON questions to gather user input for software design tasks. It does not attempt to access sensitive files, credentials, or perform network operations.
  • [INDIRECT_PROMPT_INJECTION]: The skill includes a phase for exploring the existing codebase to understand project structure and patterns. This creates a surface where the agent could ingest malicious instructions embedded in code comments or project files.
  • Ingestion points: Phase 1 (Context Discovery) involves reading project structure and related modules from the local filesystem (SKILL.md).
  • Boundary markers: None specified in the instructions to distinguish between trusted and untrusted codebase content.
  • Capability inventory: The skill has the capability to write design documents to the docs/designs/ directory (SKILL.md, Phase 5).
  • Sanitization: No explicit sanitization or filtering of codebase content is mentioned before it is processed into design summaries.
Audit Metadata
Risk Level
SAFE
Analyzed
Sep 15, 2026, 04:13 AM
Security Audit — agent-trust-hub — feature-design-assistant