cursor-sdk
Pass
Audited by Gen Agent Trust Hub on Sep 15, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The provided blueprints establish vulnerability surfaces where untrusted data enters the agent execution loop.
- Ingestion points: In
references/patterns.md, Pattern 1 ingests external source code and PR data via pull request references, and Pattern 2 ingests external tracking data from Linear tickets. - Boundary markers: Absent. The template instructions interpolate untrusted contents directly without encapsulation tags or structural separation markers.
- Capability inventory: In
references/patterns.md, the templates equip the agents with active write tools via MCP servers, specifically@modelcontextprotocol/server-githuband Linear HTTP endpoints, enabling automated comment postings, labeling, and data updates. - Sanitization: Absent. The design processes incoming content directly from external APIs without internal escaping or verification checks.
Audit Metadata