The Agent Skills Directory

[SAFE]: The skill provides legitimate game development patterns and implementation details for narrative systems. No malicious patterns, exfiltration attempts, or unauthorized access techniques were detected.
[DYNAMIC_EXECUTION]: The Python implementation for a custom dialogue runner in references/runner.md follows security best practices for dynamic evaluation. It uses the ast (Abstract Syntax Tree) module to create a restricted, whitelist-based evaluator for dialogue conditions. This implementation specifically permits only arithmetic, comparison, and boolean operations while blocking function calls and host environment access, effectively mitigating code injection risks from dialogue data sources.
[INDIRECT_PROMPT_INJECTION]: While the skill defines a system for processing external dialogue scripts (JSON/Ink/Yarn), the provided implementation includes robust input handling. The use of a sandboxed expression walker and localized line IDs ensures that narrative data cannot influence the agent's underlying behavior or execute unauthorized commands.

dialogue-systems