No. The project is intended to reduce common AI-writing patterns, not to guarantee passage through every detection system.

• The README defines the tool as rewriting AI-sounding text so it reads like a person wrote it, using 35 patterns from Wikipedia as editorial guidance.
• It preserves the original claims and forbids inventing names, numbers, dates, quotes, or citations; it does not create a verifiable proof of human authorship.
• The project insight explicitly warns against treating it as an AI detector or as a guarantee for bypassing detection.
• The README provides no independent detector, objective score, deterministic output, or compatibility testing against any detection system.

No. It should not be used to deliver a high-risk document without human review. Humanizer prohibits fabrication, but the project insight explicitly excludes unsupervised use for high-risk content.

• The README requires every name, number, date, quote, citation, and other factual detail to come from the source or the writer.
• It checks the rewrite against the original claims after the first pass, which reduces risk but is not an independent fact-verification mechanism.
• The project insight notes that the underlying agent can still introduce semantic drift, omit qualifiers, or merge sentences incorrectly.
• Legal, medical, financial, academic-citation, and high-risk technical documents are specifically identified as unsuitable for use without human review.

Yes. Humanizer fits because it is an Agent Skill written as Markdown, so it does not require a separate service, database, or compiled runtime.

• The README says it works with “any agent that supports skills” because it is just Markdown.
• Installation is available through the Skills CLI, a Claude Code plugin, Claude Desktop upload, or manual copying of SKILL.md.
• The Claude Code plugin requires version 2.1.142 or newer, which matches your stated constraint.
• The rules and multi-stage workflow are prompt-driven, so they can be reused across agents. However, output quality still depends on each agent’s instruction-following behavior and context window.

Yes, provided that you supply the sample together with the target text. The README explicitly says the sample takes precedence over the default style rules.

• The usage section supports “2-3 paragraphs of your own writing for voice matching.”
• Humanizer follows the sample’s rhythm, word choice, punctuation, and deliberate quirks rather than forcing a uniformly polished voice.
• For personal writing, it says it keeps the writer’s style; technical and reference prose remains neutral and plain.
• The project insight also identifies a limitation: voice matching may be unstable when the sample is too short, internally inconsistent, or from a substantially different genre.

It depends. Humanizer supports file-level prose rewriting, but the README does not promise batch APIs or recursive directory processing, so suitability for an automated document set is unconfirmed.

• The file example in the Usage section supplies one path: Humanize the prose in docs/launch-post.md.
• The file-processing rule explicitly preserves code, data, frontmatter, and link targets, which addresses the structural requirement for an individual file.
• Because the project runs as a Markdown Skill inside an agent rather than as a standalone batch engine, multi-file iteration depends on the particular agent workflow.
• The README does not document batch CLI flags, concurrency behavior, retry handling, or an output-directory convention.

It depends. The README explicitly targets this Markdown workflow, but protection for complex structures still depends on the connected agent following the skill instructions correctly.

• For file input, the README says it changes “only the prose,” leaving code, data, frontmatter, and link targets alone.
• The workflow performs an initial rewrite, checks the result against 35 patterns and the original claims, and revises remaining issues.
• This makes it appropriate for language-level cleanup of AI-sounding prose.
• However, Humanizer is not an independent document-processing engine; structural protection is enforced through the agent’s behavior. The README does not document compatibility with complex templates, deeply nested Markdown, or special CMS syntax.