dust-mcp-server
Pass
Audited by Gen Agent Trust Hub on Sep 29, 2026
Risk Level: SAFEINDIRECT_PROMPT_INJECTION
Full Analysis
- [SAFE]: The skill content is purely instructional, providing a step-by-step guide and boilerplate code for creating platform integrations within the Dust environment. No malicious code or obfuscation is present.
- [INDIRECT_PROMPT_INJECTION]: The skill provides instructions for building integrations that ingest data from remote platforms like Jira or Salesforce, creating a potential surface for indirect prompt injection from external data.
- Ingestion points: Remote platform APIs (Jira, HubSpot, Salesforce, etc.) mentioned in SKILL.md.
- Boundary markers: The guide recommends rendering focused markdown output but does not mandate specific delimiters for external content.
- Capability inventory: Handlers in tools/index.ts can execute API calls for list, get, and create operations on remote platforms.
- Sanitization: The guide includes a best practice to validate external API responses with Zod to ensure data integrity and catch unexpected payloads.
- [EXTERNAL_DOWNLOADS]: The documentation references official resources from the Dust domain and the Model Context Protocol website. These are well-known and appropriate technical references for the skill's purpose.
- [CREDENTIALS_UNSAFE]: The implementation guide for OAuth correctly instructs developers to manage sensitive credentials like client IDs and secrets via environment variables, following standard security practices.
Audit Metadata