convex-agents

• [INDIRECT_PROMPT_INJECTION]: The researchTopic workflow example in SKILL.md demonstrates a pattern that is susceptible to indirect prompt injection.
• Ingestion points: Data is ingested from the knowledge base (the documents table) via the agent.search call in convex/workflows.ts.
• Boundary markers: The retrieved content is joined using a simple string separator (`---

`) before being interpolated into the LLM prompt, which provides weak isolation between the system instructions and the untrusted data.

• Capability inventory: The agent is configured with tools like createTask and searchKnowledge, which could be targeted by instructions hidden in the knowledge base documents.
• Sanitization: The provided code snippets do not include logic for sanitizing or filtering document content before it is passed to the LLM.
• [EXTERNAL_DOWNLOADS]: The skill documentation references the @convex-dev/agent package and official Convex documentation domains. It also provides examples for installing the ai and openai libraries from standard package registries.