The Agent Skills Directory

[SAFE]: The skill serves as an instructional guide for narrative design and dialogue system implementation. All code snippets provided are for user implementation and do not execute during the skill's operation.
[DYNAMIC_EXECUTION]: Included in references/runner.md is a reference implementation for a sandboxed expression evaluator. This example utilizes the ast module to safely process narrative logic without resorting to eval(), representing a robust security pattern for data-driven applications.
[INDIRECT_PROMPT_INJECTION]: The skill architecture is designed to handle narrative data from external sources.
Ingestion points: The DialogueRunner in references/runner.md accepts a JSON graph containing expressions in "set" and "if" fields.
Boundary markers: The design assumes developer-authored content; no explicit content delimiters are specified in the templates.
Capability inventory: The runner manages internal state via a variable dictionary and controls UI output.
Sanitization: The evaluate function implements a secure whitelist of operations using the ast module, mitigating code injection risks from dialogue scripts.

dialogue-systems