The Agent Skills Directory

[SAFE]: The skill provides educational content and code patterns for dialogue systems with no security risks detected.
[SAFE]: Evaluated the Python expression runner in references/runner.md. The implementation uses the ast module with a strict whitelist (literals, names, arithmetic, and boolean operations) to safely process dialogue logic. It explicitly prevents function calls and host environment access, mitigating risk from dynamic expression evaluation.
[SAFE]: Evaluated the data ingestion surface for indirect prompt injection. The skill processes structured dialogue data (JSON) and provides clear boundaries between narrative content and game logic. 1. Ingestion points: JSON graph files in references/runner.md. 2. Boundary markers: Data is encapsulated in specific JSON fields like line and choices. 3. Capability inventory: Basic arithmetic and logic comparisons via the whitelisted evaluator. 4. Sanitization: The skill recommends static validation of all node targets and variables before execution, along with a whitelisted interpreter instead of eval() or exec().

dialogue-systems