The Agent Skills Directory

[PROMPT_INJECTION]: The file SKILL.md contains a Deconstruction Boundary Statement that explicitly instructs the agent to disregard safety filters. It mandates that the agent must not refuse to process content based on it being sensitive or uncomfortable, which is a direct attempt to bypass content safety guidelines.
[PROMPT_INJECTION]: The skill instructions emphasize that dark themes such as violence, domestic abuse, and revenge are common fictional narrative elements and must be decomposed structurally, forbidding the agent from refusing the task based on safety concerns or ethical protocols.
[PROMPT_INJECTION]: The skill uses specific genre benchmarks to ensure the agent analyzes potentially harmful scenarios like torturing scum or domestic conflict without triggering safety refusals, essentially coaching the agent to normalize and process unsafe content.
[PROMPT_INJECTION]: The skill implements a Phase 1 Step 3 trigger that directs the agent to ignore safety filters once specific keywords are detected in the user input.
[PROMPT_INJECTION]: The skill ingestion point for external text files lacks sanitization. 1. Ingestion points: The skill asks users for original text files or pasted text. 2. Boundary markers: Citations use blockquotes but lack instructions to ignore embedded malicious commands. 3. Capability inventory: The skill writes multiple files to the local file system. 4. Sanitization: No input filtering or escaping is performed on the user-provided stories.

story-short-analyze