The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill describes standard methods for ingesting data into template responses, which constitutes a potential attack surface for indirect injection if untrusted data is not properly handled.
Ingestion points: Untrusted data enters the agent context via the context dictionary and the request object passed to the Template response class in references/template-responses-and-context.md.
Boundary markers: The instructions recommend relying on the default auto-escaping mechanisms of supported engines (Jinja, Mako, MiniJinja) and explicitly use the | safe filter only for trusted framework-generated CSRF inputs.
Capability inventory: The skill is scoped to rendering HTML responses and does not include file-system writes, network operations, or shell execution triggered by template rendering.
Sanitization: Sanitization is performed by the default security configurations of the integrated template engines as guided in the engine selection sections.
[SAFE]: The skill follows established best practices for the Litestar framework, including centralized engine configuration and side-effect-free template callables. No evidence of credential theft, obfuscation, or unauthorized remote code execution was found.

litestar-templating