The Agent Skills Directory

[PROMPT_INJECTION] (LOW): The chat application example in references/chat-app-example.md is vulnerable to Indirect Prompt Injection.
Ingestion points: User-provided content is ingested via the sendMessage mutation in convex/index.ts and stored in the database.
Boundary markers: Absent. In the loadContext function, user messages are interpolated into strings (e.g., user.name: message.content) and passed directly to the OpenAI API without delimiters or instructions to ignore embedded commands.
Capability inventory: The skill has database write capabilities via ctx.runMutation and can perform external network requests to OpenAI.
Sanitization: Absent. The implementation does not filter, escape, or validate the user-generated content before processing it with the LLM.
[EXTERNAL_DOWNLOADS] (SAFE): The skill references standard, well-known software packages in its example configuration.
Evidence: package.json specifies convex and openai (a trusted organization).

convex