Humanizer.Local.Backup

This skill appears to do what it says (humanize text) and doesn't request credentials or install software, but there are a few practical risks to consider before enabling it: - Limit scope explicitly: when invoking, tell the agent to operate only on the provided text and to not read or search other files or repositories unless you explicitly authorize that. The allowed-tools list (Read/Write/Grep/Glob) could enable broader file access if not constrained. - Avoid secrets or sensitive material: do not feed confidential data (passwords, private keys, proprietary facts) into the skill unless you are comfortable with the agent having read/write access to those documents. - Watch for invented claims and personalization: the skill instructs the agent to 'add soul' and use first‑person voice, which can cause the model to add opinions or assert facts not in the source. Carefully review outputs for hallucinated or inaccurate statements before publishing. - Test with non-sensitive examples first: verify how it behaves when given only inline text vs. when asked to operate on files; request explicit confirmation of which files (if any) the skill will read. If you need stronger guarantees, ask the skill author (or your platform admin) to restrict allowed-tools so the skill can only read the specific input provided and cannot run Grep/Glob over the workspace.