The Agent Skills Directory

[SAFE]: The skill consists entirely of natural language instructions and structural templates for text analysis and rewriting. There are no command-line executions, remote script downloads, or network-enabled operations described in the instructions.
[INDIRECT_PROMPT_INJECTION]: As a text-processing utility, the skill inherently ingests untrusted user data for rewriting. However, the skill does not utilize any dangerous tools (e.g., shell access, file writing, or API calls) that could be misused if a user provided a malicious prompt embedded in their text. The risk is limited to the agent potentially following instructions within the text rather than rewriting it, which is a standard LLM behavior and not a vulnerability in the skill itself.
[METADATA_POISONING]: The metadata fields (name and description) are consistent with the skill's stated purpose and contain no hidden instructions or deceptive content.

de-ai-writing