The Agent Skills Directory

[PROMPT_INJECTION]: The skill demonstrates an indirect prompt injection surface in its code examples for both TypeScript and Python.
Ingestion points: Untrusted data from event.data.content (in SKILL.md) and ctx.event.data.text (in references/dependency-injection.md) is ingested from event payloads into the agent context.
Boundary markers: The examples lack delimiters or 'ignore embedded instructions' warnings when interpolating untrusted data into prompts.
Capability inventory: The skill demonstrates the ability to invoke the openai client to process this untrusted data via openai.chat.completions.create.
Sanitization: No sanitization or validation logic is shown to filter or escape the external content before it is processed by the model.
[COMMAND_EXECUTION]: The documentation includes standard installation instructions for official vendor packages, such as @inngest/middleware-encryption and @inngest/middleware-sentry. These resources originate from the vendor's own namespace and are documented for their intended purpose.
[CREDENTIALS_UNSAFE]: The skill follows secure practices for credential management, consistently demonstrating the use of environment variables (e.g., process.env.ENCRYPTION_KEY, os.environ["OPENAI_API_KEY"]) and placeholders rather than hardcoding sensitive secrets.

inngest-middleware