The Agent Skills Directory

[PROMPT_INJECTION]: The skill instructions are focused on guiding the agent through a creative writing process. There are no patterns suggesting attempts to bypass safety filters or override system-level instructions for malicious purposes. The internal directives such as 'Must comply' and 'Prohibited items' are standard constraints for maintaining output quality.
[DATA_EXPOSURE]: No hardcoded credentials, sensitive file paths, or network exfiltration patterns were identified. The skill operates entirely on the input provided by the user and the included markdown documentation.
[OBFUSCATION]: The content across all four files consists of plain-text markdown and Chinese characters. No Base64, zero-width characters, homoglyphs, or encoded strings were found.
[REMOTE_CODE_EXECUTION]: The skill does not perform any remote code execution or package installations. It relies on internal logic and text generation rather than external scripts or binaries.
[PRIVILEGE_ESCALATION]: No commands related to privilege escalation (e.g., sudo, chmod) or persistent mechanisms were found.
[INDIRECT_PROMPT_INJECTION]: While the skill ingests user-provided viewpoints, it lacks dangerous capabilities such as file-system writes, network access, or shell execution that could be exploited via injection. The risk surface is limited to text generation.
[DYNAMIC_EXECUTION]: The skill does not utilize any dynamic loading, runtime compilation, or unsafe deserialization. It references local markdown files (@4d-framework.md, etc.) for methodology instructions.

social-polish