design-memory
Pass
Audited by Gen Agent Trust Hub on Oct 5, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill establishes a framework for reading and writing project design decisions to a
.sumi/directory in JSON format. This creates a surface where external data (design decisions) is ingested into the agent's context, potentially allowing for indirect prompt injection if data values are interpreted as instructions. - Ingestion points: Structured JSON files located in the
.sumi/directory, specificallystyle.json,brief.json,map.json,vision.json,wireframe-<screen>.json, anddecisions.logas defined inSKILL.md. - Boundary markers: The skill enforces a strict canonical schema and merge rules which act as a structural boundary, though it does not explicitly define escape sequences for string values.
- Capability inventory: None. This skill is a contract definition and does not include any scripts, tools, or executable commands.
- Sanitization: The skill relies on rigid schema validation (e.g., via
scripts/check-corpus.py) to ensure data integrity and prevent unexpected top-level keys.
Audit Metadata