The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents the creation of middleware that intercepts and processes event payloads and function results, establishing a potential attack surface for indirect prompt injection.
Ingestion points: Untrusted data enters the agent context through middleware hooks like transformInput (accessing ctx.event.data) and transformOutput (accessing result.data) as shown in SKILL.md, references/built-in-middleware.md, and references/dependency-injection.md.
Boundary markers: None are present in the provided examples to delimit external content or instruct the model to ignore embedded instructions.
Capability inventory: The documented middleware patterns possess capabilities for network operations (fetch calls in reportError), database interactions (via db client), and external logging/tracing (Sentry integration).
Sanitization: The examples do not demonstrate explicit sanitization, validation, or filtering of the processed event data before it is logged or sent to external services.
[EXTERNAL_DOWNLOADS]: The skill references several external packages and services necessary for the described middleware functionality. These are documented as follows:
Recommends installation of official vendor packages: @inngest/middleware-encryption and @inngest/middleware-sentry.
Recommends well-known service clients and libraries: @sentry/node, openai, stripe, @prisma/client, and redis.
[COMMAND_EXECUTION]: The documentation includes standard shell commands for package management (e.g., npm install) to facilitate the setup of the described middleware components.

inngest-middleware