The Agent Skills Directory

[SAFE]: A comprehensive review of the 11 files comprising this skill (including orchestrator, agent instructions, and domain references) revealed no malicious behaviors or security risks. The skill adheres strictly to its stated purpose of text humanization.
[PROMPT_INJECTION]: The skill employs extensive instructional constraints to identify and remove AI 'telltales' (e.g., zero tolerance for em dashes, rhetorical questions, and specific vocabulary clusters). These instructions are legitimate content editing rules and do not involve malicious prompt engineering aimed at bypassing agent safety filters or extracting internal system prompts.
[DATA_EXFILTRATION]: No unauthorized file access or credential harvesting patterns were detected. The skill reads local project context (e.g., 'research/product-context.md') to align voice injection with the user's brand, which is a standard functional requirement.
[REMOTE_CODE_EXECUTION]: The skill does not download external scripts or execute untrusted code. It uses a structured multi-agent dispatch protocol based on local markdown instructions and platform-provided tools.
[COMMAND_EXECUTION]: Although the 'Bash' tool is listed in the skill's allowed-tools frontmatter, no dangerous or arbitrary shell command patterns were found within the agent instructions. All operations remain within the scope of text processing and local file management.

humanize