The Agent Skills Directory

[PROMPT_INJECTION]: The skill's instructions are focused entirely on German copywriting and grammatical consistency. It contains no commands to override safety filters, disregard previous instructions, or adopt restricted personas.
[DATA_EXPOSURE]: The skill reads from a local brand-brief.md file to gather project context. This is a standard and safe practice for AI agents to align with project requirements. No hardcoded credentials or access to sensitive system paths (e.g., .ssh, .aws) were found.
[REMOTE_CODE_EXECUTION]: The skill does not perform any network operations, package installations, or shell command executions. It is a text-transformation skill that runs entirely within the LLM's dialogue context.
[INDIRECT_PROMPT_INJECTION]: While the skill processes untrusted user input (German copy), it lacks the capabilities (such as file writing, network access, or subprocess execution) that would make an indirect injection attack viable. The risk is negligible as the agent's output is limited to copy suggestions.
[OBFUSCATION]: The instructions and referenced materials are written in clear, plain text. No Base64, hex encoding, zero-width characters, or other concealment techniques were detected.

du-sie-check