The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill establishes patterns for ingesting untrusted data from external sources, which creates a potential surface for indirect prompt injection attacks.
Ingestion points: Untrusted data enters the agent context through httpAction handlers in convex/http.ts (e.g., req.json()) and via external API responses in convex/ai.ts (e.g., openai.chat.completions.create).
Boundary markers: The provided code examples lack explicit delimiters or instructions to the agent to ignore potentially malicious embedded commands within the processed data.
Capability inventory: The skill utilizes ctx.runMutation, ctx.runQuery, and ctx.db.insert across its scripts, allowing ingested data to trigger database modifications or further function calls.
Sanitization: Although the skill mandates the use of Convex validators for type safety, these do not perform content-level sanitization to prevent adversarial instructions from influencing subsequent agent steps.

convex-actions-general