The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill documents several mechanisms for ingesting external, untrusted data that is then used within the agent's context. Specifically:
Ingestion points: Data enters the system through inbound webhooks (webhooks.md), Segment event triggers (journeys.md), and HTTP responses merged via journey fetch nodes (journeys.md, batching.md).
Boundary markers: The provided documentation and code snippets do not emphasize the use of delimiters or specific safety instructions to isolate untrusted data when it is interpolated into templates using {{data.field}} or {{profile.field}} syntax.
Capability inventory: The platform supports powerful capabilities that can be triggered by this data, including making network requests via fetch nodes and executing LLM prompts via ai nodes (journeys.md).
Sanitization: Explicit guidance on sanitizing or escaping external content before interpolation is missing, relying on the platform's default handling.
[DYNAMIC_EXECUTION]: The Elemental templating language includes features that execute logic at runtime:
Conditional Logic: The if property on elements evaluates JavaScript expressions against message data (elemental.md).
Jsonnet: The jsonnet element allows programmatic content generation by evaluating Jsonnet source code at render time (elemental.md).
These features are part of the platform's core functionality but represent a dynamic execution surface that operates on potentially untrusted input.
[EXTERNAL_DOWNLOADS]: The skill provides instructions for installing official Courier tools and libraries:
CLI: Recommends npm install -g @trycourier/cli and points to the vendor's GitHub releases (github.com/trycourier/courier-cli).
SDKs: Directs users to install @trycourier/courier via npm and trycourier via pip.
UI Components: Instructions for installing React, Vue, Angular, and React Native components from the @trycourier scope on npm.

courier