The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill processes untrusted external data that could contain malicious instructions designed to influence the agent's behavior.
Ingestion points: The render method in the Template class and the Many combinator ingest a ctx (context) dictionary containing external data.
Boundary markers: The DSL does not provide built-in mechanism for prompt delimiters or 'ignore instructions' markers when interpolating context.
Capability inventory: The skill's primary purpose is to generate structured prompt sequences for consumption by an LLM.
Sanitization: The Template.render method uses standard Python string formatting (self.template_str.format(**ctx)) without escaping or validating the content of the context variables.

prompt-dsl