Humanizer: A Skill to Humanize Text for Claude Code and OpenCode

1. Detailed Overview

• The Humanizer skill is designed for Claude Code and OpenCode and serves as a text transformation tool that reduces recognizable AI-generated writing signatures. Its goal is to produce text that reads more naturally and sound more human, aligning with human writing rhythms, word choices, and stylistic quirks.
• At its core, the skill relies on a controlled rewrite process informed by a comprehensive set of patterns identified in AI-generated prose. It does not simply “clean” text; it analyzes sentence rhythm, diction, and typical AI-isms, then applies targeted adjustments to produce more natural output.
• The approach includes an optional, explicit voice calibration workflow. By ingesting a user-provided sample of their own writing, the system can tailor rewrites to reflect personal style rather than producing generic cleanups. This makes the result feel more like the author rather than a generic, machine-generated rewrite.
• A notable feature of the design is a final audit pass labeled “obviously AI generated.” After the initial rewrite, the skill applies a secondary review to catch lingering AI markers and refine phrasing further. This two-pass approach helps reduce artifacts that might still hint at automation.
• The project draws on an established research perspective summarized in the guide “Signs of AI writing.” The core insight is that large language models (LLMs) rely on statistical patterns to predict what comes next, which tends to produce output that aligns with the broad statistical likelihood across many contexts. The Humanizer tool uses this understanding to identify patterns that typically betray AI authorship and address them directly in the rewrite.
• The skill’s design emphasizes practical, observable outputs. It offers straightforward usage through simple commands in both Claude Code and OpenCode environments, plus optional prompts that invite direct humanizing requests. The intention is to provide a reliable, repeatable workflow for users who want more human-sounding text without investing in bespoke stylistics for each piece.

1. Installation

• The installation process is designed for two distinct platforms, with a shared philosophy of compatibility so a single clone can serve both tools under certain configurations.
• For Claude Code:
• Clone directly into Claude Code’s skills directory: create the appropriate directory path and pull the repository contents into ~/.claude/skills/humanizer.
• Alternative manual path: if you already have the repo cloned elsewhere, copy the skill file (SKILL.md) into ~/.claude/skills/humanizer/ to integrate the skill without re-cloning.
• For OpenCode:
• Clone into OpenCode’s skills directory: create the necessary directory on ~/.config/opencode/skills and clone the Humanizer repository there.
• Alternative manual path: copy SKILL.md into ~/.config/opencode/skills/humanizer/ if you have the repo present locally.
• Compatibility note:
• OpenCode also scans ~/.claude/skills/ for compatibility. Therefore, a single clone placed in either path will often be usable across both tools. This cross-compatibility means users can streamline setup by deploying once and sharing across environments when possible.

1. Quick Start and Usage

• Claude Code command:
• To humanize text, enter the command: /humanizer [paste your text here]
• This initiates the rewrite process, applying the AI-signature reduction principles to the pasted material.
• OpenCode command:
• The OpenCode workflow mirrors Claude Code: /humanizer [paste your text here]
• The rewrite is produced with the same objective of yielding a more natural, human-sounding version.
• Inline prompting option:
• If you prefer conversational prompts, you can ask the model directly: “Please humanize this text: [your text]”
• This prompt approach leverages the model’s capabilities to interpret your intent and perform the transformation in-line.
• Voice calibration workflow:
• To calibrate to your personal style, provide a sample of your own writing using:
  • /humanizer Here's a sample of my writing for voice matching: [paste 2-3 paragraphs of your own writing]
  • Then apply the command to your text: Now humanize this text: [paste AI text to humanize]
• The outcome adopts your sentence rhythm, preferred word choices, and distinctive quirks rather than defaulting to a generic, neutral tone.
• Result expectations:
• The humanized output should read as a more natural prose style, with reduced reliance on common AI-flagging constructs such as overly promotional phrasing, vague attributions, or repetitive patterns found in AI-synthesized text.
• The two-pass approach (initial rewrite plus a final audit) helps catch residual AI-isms and produce a smoother final draft.

1. Conceptual Foundation and Key Insight

• The Humanizer tool is grounded in a practical interpretation of AI-writing indicators. It uses a set of detection-driven patterns to guide how it rewrites text, focusing on the kinds of stylistic choices that tend to betray machine authorship.
• The guiding principle cited from well-known discussions of AI writing is that LLMs generate text by selecting statistically likely continuations. The Humanizer’s strategy is to shift away from those high-likelihood patterns toward phrasing that better reflects human authorship, including the idiosyncrasies, informal rhythms, and contextual nuance typical of individual writers.
• This approach is augmented by an explicit audit step designed to identify lingering AI cues after the rewrite. The audit serves as a safety net to ensure the final product does not exhibit common AI-signal markers.

1. 29 Patterns Detected: A Structured Palette of Transformations

• The system organizes its set of insights into multiple pattern groups, each with concrete before-and-after illustrations that guide where and how changes occur.
• Content Patterns:
• Pattern 1: Significance inflation
  • Before: “marking a pivotal moment in the evolution of…”
  • After: “was established in 1989 to collect regional statistics”
• Pattern 2: Notability name-dropping
  • Before: “cited in NYT, BBC, FT, and The Hindu”
  • After: “In a 2024 NYT interview, she argued…”
• Pattern 3: Superficial -ing analyses
  • Before: “symbolizing… reflecting… showcasing…”
  • After: Remove or expand with actual sources
• Pattern 4: Promotional language
  • Before: “nestled within the breathtaking region”
  • After: “is a town in the Gonder region”
• Pattern 5: Vague attributions
  • Before: “Experts believe it plays a crucial role”
  • After: “according to a 2019 survey by…”
• Pattern 6: Formulaic challenges
  • Before: “Despite challenges… continues to thrive”
  • After: “Specific facts about actual challenges”
• Language Patterns:
• Pattern 7: AI vocabulary
  • Before: “Actually… additionally… testament… landscape… showcasing”
  • After: “also… remain common”
• Pattern 8: Copula avoidance
  • Before: “serves as… features… boasts”
  • After: “is… has”
• Pattern 9: Negative parallelisms / tailing negations
  • Before: “It's not just X, it's Y”, “…, no guessing”
  • After: State the point directly
• Pattern 10: Rule of three
  • Before: “innovation, inspiration, and insights”
  • After: Use natural item count without forcing a parallel list
• Pattern 11: Synonym cycling
  • Before: “protagonist… main character… central figure… hero”
  • After: Choose the clearest single term to preserve clarity
• Pattern 12: False ranges
  • Before: “from the Big Bang to dark matter”
  • After: List topics directly with specificity
• Pattern 13: Passive voice / subjectless fragments
  • Before: “No configuration file needed”
  • After: Name the actor when it helps clarity
• Style Patterns:
• Pattern 14: Em dash overuse
  • Before: “institutions—not the people—yet this continues—”
  • After: Prefer commas or periods
• Pattern 15: Boldface overuse
  • Before: “OKRs, KPIs, BMC”
  • After: “OKRs, KPIs, BMC”
• Pattern 16: Inline-header lists
  • Before: “Performance: Performance improved”
  • After: Convert to prose
• Pattern 17: Title Case Headings
  • Before: “Strategic Negotiations And Partnerships”
  • After: “Strategic negotiations and partnerships”
• Pattern 18: Emojis
  • Before: “🚀 Launch Phase: 💡 Key Insight:”
  • After: Remove emojis
• Pattern 19: Curly quotes
  • Before: “ said “the project” ”
  • After: “said “the project”” (consistent typographic style)
• Pattern 26: Hyphenated word pairs
  • Before: “cross-functional, data-driven, client-facing”
  • After: Drop hyphens on common word pairs unless needed for clarity
• Persuasive Authority and Signposting Patterns:
• Pattern 27: Persuasive authority tropes
  • Before: “At its core, what matters is…”
  • After: State the point directly with supporting details
• Pattern 28: Signposting announcements
  • Before: “Let's dive in”, “Here's what you need to know”
  • After: Start with the content and then guide the reader
• Fragmented Headers and Communication Patterns:
• Pattern 29: Fragmented headers
  • Before: “Performance” + “Speed matters.”
  • After: Let the heading carry the weight of the section with the prose following
• Other Pattern Notes:
• Filling and hedging considerations (see below)
• The system also addresses direct “chatbot artifacts” and cutoff disclaimers, ensuring the rewrite remains focused and avoids meta-text that could expose AI authorship.
• Practical takeaway:
• Each pattern is designed to be actionable. The user benefits from more natural phrasing, clearer attribution, and more direct statement of facts, while still preserving the intended meaning and content.

1. Full Example: Before and After Narratives

• Before (AI-sounding) narrative:
• The example begins with a polished but generic introduction to AI-assisted coding, highlighting speed, quality, and adoption in broad terms. It cites media outlets and discusses the supply chain of automation, tests, and documentation generation as universal benefits. It uses promotional phrasing and a generalized, sweeping description of impact across industries.
• After (Humanized) narrative:
• The rewritten narrative presents a more grounded view. It emphasizes practical benefits like speeding up boilerplate tasks and drafting tests, while noting that human review remains essential. It cautions against overreliance on automatic suggestions, citing potential misalignment with project goals and the need for robust testing. The rewrite foregrounds real-world constraints and a balanced perspective, with concise, readable prose that avoids grandiose claims and generic hype.
• The two-pass process ensures the final result feels both human and precise, emphasizing the user’s voice and the concrete realities of software development work.

1. Practical Workflows and Best Practices

• Usage workflow:
• Install the Humanizer skill in your preferred environment using the stated paths.
• Decide whether to run via direct command or via a prompt that explicitly asks for humanization.
• If you require alignment with your personal voice, provide a short but representative sample of your own writing and then apply the transformation to your target text.
• Use the final audit pass to ensure no lingering AI patterns remain, iterating if necessary.
• Content strategy:
• Rely on the voice calibration step for consistent results across multiple documents.
• Combine the tool with a human editor for critical texts to maximize reliability and nuance.
• When working with sensitive material, consider running additional reviews to ensure accuracy and tone alignment.
• Platform considerations:
• The compatibility note means users can streamline maintenance by hosting a single clone for both Claude Code and OpenCode environments.
• If you switch between tools or environments, verify that the skill path in your configuration remains accessible and correctly named (for example, the humanizer directory under the relevant skills path).

1. Version History and Evolution

• Version 2.5.1
• Added a passive-voice / subjectless-fragment rule, bringing the total pattern count to 29.
• Version 2.5.0
• Expanded patterns for persuasive framing, signposting, and fragmented headers.
• Extended negative parallelisms to cover tailing negations.
• Tightened wording around em dash overuse and updated frontmatter wording to emphasize “filler phrases.”
• Version 2.4.0
• Introduced voice calibration to match the user’s personal writing style from samples.
• Version 2.3.0
• Added pattern #25: hyphenated word pair overuse (e.g., cross-functional, data-driven) and recommended hyphen handling.
• Version 2.2.0
• Implemented a final “obviously AI generated” audit plus a second-pass rewrite prompt to improve detection and resolution of AI traits.
• Version 2.1.1
• Fixed a pattern example discrepancy (curly quotes vs straight quotes) to ensure consistent demonstrations.
• Version 2.1.0
• Added before/after examples for all 24 patterns, expanding coverage and clarity.
• Version 2.0.0
• Rewrote the entire framework based on raw content inspiration from a Wikipedia article on AI writing signs.
• Version 1.0.0
• Initial release, establishing the core capabilities and workflow.

1. Licensing

• The Humanizer project is released under the MIT license, which governs how the skill may be used, modified, and redistributed.
• The license enables integration into Claude Code and OpenCode environments, as well as adaptation for personal or organizational workflows, subject to the terms of the MIT license.

1. Practical Summary: What You Get

• A robust, two-pass rewriting workflow that targets AI-writing signs without sacrificing meaning or clarity.
• A voice calibration feature that helps align the rewritten text with your own distinctive style.
• A comprehensive set of 29 patterns spanning content, language, style, and communication norms, each with explicit Before/After guidance to illustrate the intended transformation.
• Simple, consistent usage across Claude Code and OpenCode via familiar commands, with cross-compatibility considerations that make setup efficient.
• An explicit audit pass designed to minimize residual AI signatures and ensure the rewritten text reads as naturally human.

1. Final Thoughts

• The Humanizer skill combines practical engineering with an evidence-informed approach to human-like writing. By addressing a wide spectrum of AI-writing traits—from inflated significance and vague attributions to overused promotional tropes and stylistic quirks—it provides a structured, repeatable method for producing text that feels authentically human while preserving the author’s intent.
• For users who frequently draft content that requires a tone of natural fluency—ranging from reports and briefs to blog posts and communications—the tool offers a disciplined path to cleaner, more readable prose without surrendering nuance.
• While it is a powerful assistant, the skill intentionally positions human judgment as the final arbiter. The two-pass methodology and voice calibration option are designed to keep humans in the loop, ensuring that the final output matches the writer’s goals, intent, and voice.

Note: This description preserves the core content and structure of the input, reframing it into a detailed, user-facing guide that emphasizes installation, usage, patterns, and practical workflows. There are no tables in this description, and it follows a structured format with numbered sections and bullet points, suitable for a thorough, human-readable reference.