The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill defines a surface for processing user-provided game descriptions to generate roleplay matrices. However, there are no exploitable capabilities such as file system writes, network requests, or subprocess execution, meaning the ingestion of untrusted game data poses no risk to the host environment.
[DATA_EXFILTRATION]: Analysis confirmed that terms like 'permissions', 'authorization', and 'access history' refer exclusively to fictional game mechanics (e.g., keys to a medicine cage) and not to system-level credentials or sensitive user data. No network operations were detected.
[OBFUSCATION]: No evidence of Base64 encoding, zero-width characters, homoglyphs, or other hidden content techniques was found within the skill documentation, templates, or examples.
[COMMAND_EXECUTION]: The skill does not utilize shell commands, script generation, or dynamic execution patterns. Mentions of 'runtime implementation' refer to the logical consistency of a game's state rather than the execution of code on the agent's platform.
[METADATA_POISONING]: The skill's metadata and instructional goals are consistent. The safety instructions in Section 7 and the 'Why identity fails' reference file explicitly prohibit using protected characteristics as markers for danger, reinforcing safe behavior.

worldview-game-horror-role-and-identity-pressure