convex-agent
Pass
Audited by Gen Agent Trust Hub on Oct 6, 2026
Risk Level: SAFEEXTERNAL_DOWNLOADSINDIRECT_PROMPT_INJECTION
Full Analysis
- [EXTERNAL_DOWNLOADS]: The skill instructs the user to install the
@convex-dev/agentpackage, which is the official library for Convex AI agent integration. - [INDIRECT_PROMPT_INJECTION]: The skill defines a RAG workflow involving the processing of external documents and message history, which creates a surface for indirect prompt injection.
- Ingestion points: Ingests external documents for vector indexing and retrieves them during tool calls; maintains persistent message history in durable threads (SKILL.md).
- Boundary markers: The high-level instructions do not specify explicit delimiters or guardrails for handling untrusted content within ingested data.
- Capability inventory: Involves LLM model invocation, tool calling, and vector database search.
- Sanitization: No specific sanitization or filtering logic is outlined in the workflow provided.
Audit Metadata