Skills

RayCodes_Humanizer: A Skill That Locally Rewrites AI Technical Docs into Conversational Blogs

Want to turn locally generated AI technical docs into a natural, conversational developer blog? RayCodes_Humanizer is the right fit; you must install Ollama and register the noema2b model first, and it only accepts Markdown input.

Official source · RayCodes_Humanizer

Install

npx skills add blader/humanizer --global

Who It's For

  • For anyone who needs to rewrite AI-generated technical documentation into a natural, conversational developer blog: project authors, technical writers, and teams publishing internal docs. It emphasizes fully preserving code blocks, command lines, and links during rewriting.
  • It runs locally and depends on Ollama, and is primarily aimed at Markdown input (the officially specified input file is inputs/input_doc.md). If your text isn't Markdown, or you can't install Ollama and a Python environment locally, the official evidence offers no alternative — this is unknown territory.
  • Its rewriting style is positioned as "conversational developer blog," so it's not suited for scenarios that require preserving a formal written style. The official docs don't state whether other writing styles are supported — don't assume you can switch styles.

First Try in 3 Minutes

The officially recommended installation method is the Skills CLI (Option A, Recommended). Prerequisites: Ollama installed locally, and the backend model registered as officially recommended — noema2b (or llama3.2:3b). The official model registration process installs dependencies and creates a Modelfile:

pip install urllib3 ; echo "FROM hf.co/NoemaAI-labs/Noema-2B-GGUF\nPARAMETER temperature 0.7\nPARAMETER presence_penalty 1.5" > Modelfile ; ollama create noema2b -f Modelfile

Then run the official command in an agent that supports Skills CLI:

npx skills add blader/humanizer --global

For your first try, use a disposable Markdown draft: place it in the officially specified input file inputs/input_doc.md, then run the engine:

python main.py

Success signal: the official docs state the rewritten result is generated at outputs/outputs.md, along with an exact generation timestamp. Open that file and check that code blocks, command lines, and links are fully preserved. Note: running the engine means executing python main.py — an operation that reads your input document, calls the local Ollama model, and writes an output file, so confirm beforehand that the input content is safe to process. Whether the skill also reads/writes other files, executes other commands, or accesses the network is not stated in the official evidence — treat this as unknown, and don't assume it's completely side-effect free.

Another officially supported installation path is the Claude Code plugin marketplace: /plugin marketplace add blader/humanizer, followed by /plugin install humanizer@humanizer. This is an alternative path, not the main recommendation in this article.

Use Cases

Use case 1: Polishing an AI-generated project README. Copy the dense, machine-flavored README into inputs/input_doc.md, run python main.py, then compare against outputs/outputs.md to see the conversational version with boilerplate and corporate tone removed. Focus on verifying that code blocks and install commands are preserved verbatim — the result can be used directly as blog material.

Use case 2: Rewriting an AI-written technical proposal into a team sharing document. The official docs state the output is a "human-readable blog post or developer conversation." You can keep the command-line examples from your proposal in the input document, then after running, check whether the output file has colloquialized the narrative while fully preserving commands and links. The official evidence only confirms the input → run → output file flow; finer-grained segmentation or style-control parameters are not documented — don't assume they exist.

Usage Boundaries

  • Files and commands: Running python main.py reads inputs/input_doc.md and writes outputs/outputs.md; whether it accesses other files, executes other commands, or goes online is not stated officially — unknown.
  • Runtime environment: Requires local Ollama (recommended noema2b:latest or llama3.2:3b) and Python 3.12 for the core logic. Before first use, confirm the model has been registered successfully with the official commands.
  • Installation differences: Skills CLI and the Claude Code plugin marketplace are two distinct official paths for different environments — don't mix them.
  • Unconfirmed information: Supported input languages, whether output style is configurable, and how invocation is triggered in specific clients are all not covered by the official evidence — verify on a small scale before relying on it.