context-engineering
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [INDIRECT_PROMPT_INJECTION]: The skill describes the practice of loading external data (e.g., third-party API responses, user-submitted content, and external documentation) into the agent's context, which creates an indirect prompt injection attack surface.
- Ingestion points: Identified in the 'Level 3: Relevant Source Files' section, which suggests reading third-party API responses and user-submitted content.
- Boundary markers: The skill explicitly advises treating instruction-like content as data to be surfaced to the user rather than followed as directives.
- Capability inventory: The instructions mention the use of Model Context Protocol (MCP) servers providing access to the filesystem, PostgreSQL databases, and GitHub repositories.
- Sanitization: It recommends manual verification and treating external documentation as 'untrusted' until reviewed by a user.
- [SAFE]: The skill contains no executable malicious code, hardcoded secrets, or hidden payloads. The listed commands (e.g., npm test, npm run dev) are provided as illustrative examples for project rules files and do not perform unauthorized actions.
Audit Metadata