Humanizer

This skill appears to do what it says — editing text to remove AI-like phrasing — but there are a few cautionary points before installing: - Allowed tools are broad: the SKILL.md permits Read/Write/Edit/Grep/Glob. While editing a supplied text is fine, these permissions would let the agent search the filesystem and modify files if it chooses. If you install it, restrict its tool permissions or ensure the agent only processes text you provide. - Metadata inconsistencies: versions and owner IDs differ across package.json, SKILL.md, and _meta.json, and no homepage or source is listed. That doesn't prove malice, but it makes provenance harder to verify. Prefer skills with clear, matching metadata and a known source. - No network installs or credential requests were found (good), but because this is instruction-only the agent's behavior depends entirely on the platform's enforcement of allowed-tools and on how the skill is invoked. If you need to be cautious, run it in a sandboxed agent or ask the skill author to confirm that it will only operate on user-supplied text and will not perform filesystem searches or automated edits outside the target input. If you want higher confidence before use, ask the publisher to: (1) unify metadata (version/owner), (2) remove Grep/Glob if not needed, or (3) explicitly state that the skill will only read and write the provided text and not search or modify other files.