AI Humanizer for Claude, Codex, Cursor, and OpenClaw

An AI humanizer should not be a mystery form that rewrites once and asks you to trust a score. A useful one asks what kind of text you need, rewrites in passes, and checks whether the pass improved the draft.

The agentic AI Humanizer skill is that workflow for Claude Code, Codex CLI, Cursor, Hermes Agent, OpenClaw, Gemini CLI, OpenCode, and similar agent tools. Its core humanization workflow does not need Slop or Not. It runs five rewrite passes, supports saved preferences, and can match a writing sample wherever your agent can load the skill. With Slop or Not installed on Mac and Pro active, it also adds on-device AI detector scores, readability analysis, Text Cleanup, and cleanup stats.

If you landed here from search and have never used Slop or Not, this is the full setup. You can install the Slop or Not plugin repository first and start the AI humanizer without the Mac app. Add Slop or Not afterward when you want on-device AI detector scoring and the Text Cleanup summary.

What Is the AI Humanizer Skill?
What Do You Need Before You Install It?
How Do You Install the AI Humanizer Skill?
How Do You Connect Slop or Not for Local Checks?
How Do You Run the AI Humanizer Loop?
Can an AI Humanizer Bypass AI Detectors?
Privacy Notes
FAQ

What Is the AI Humanizer Skill?

The agentic AI Humanizer skill rewrites AI-generated text with a five-pass workflow. Before it edits, it asks for dialect, reading level, tone, and length. It can also learn a compact voice fingerprint from a writing sample, so the output moves toward your own rhythm instead of a generic "human" style.

The skill is not part of the Slop or Not app. It runs inside your agent. Slop or Not supplies the optional local measurement layer through the slop CLI or slop mcp server, both bundled with the Mac app for Pro users.

With Pro active, the measured loop uses three local tools:

detect_text or slop text --json for AI probability.
analyze_readability or slop readability --json for Flesch-Kincaid grade level.
clean_text or slop cleanup for hidden characters, homoglyphs, punctuation artifacts, and dialect cleanup counts.

If Slop or Not is not available, or Pro is not active, the skill still runs all five rewrite passes and can still match your voice. It reports score and grade as n/a, skips Text Cleanup stats, and avoids claiming detector convergence because no on-device AI detector ran.

What Do You Need Before You Install It?

For core AI humanization, you need an agent that can load skills or plugins. Claude Code and Codex CLI have plugin paths. Claude Code, Cursor, and Windsurf can also use npx skills add. Codex CLI, Gemini CLI, and OpenCode can load a cloned skill folder when plugins are not available.

For on-device AI detector feedback, readability checks, Text Cleanup, and cleanup stats, you also need:

A Mac with Apple silicon.
Slop or Not for Mac installed from the download page.
Pro active in the app.
Command-line access from Slop or Not Settings > Command Line, or the bundled binary path.

The app binary lives here:

/Applications/Slop Or Not.app/Contents/MacOS/slop

Open Slop or Not once before you wire it into an agent. Then open Settings > Subscription and activate Pro. The CLI and MCP server return Pro-required errors until the app sees an active purchase.

How Do You Install the AI Humanizer Skill?

Install the plugin bundle when your agent supports plugins. The bundle name is slopornot; the skill command is still /agentic-humanizer outside Claude Code plugin installs.

For Codex CLI, add the marketplace:

codex plugin marketplace add numen-tech/slopornot

Then run codex, open /plugins, switch to the slopornot marketplace, and choose Install plugin.

For Claude Code, run these commands inside Claude Code:

/plugin marketplace add numen-tech/slopornot
/plugin install slopornot@slopornot

Claude Code plugin installs namespace the command by plugin name:

/slopornot:agentic-humanizer

For a non-interactive Claude Code install, first add the marketplace from an interactive session, then run:

claude plugin install slopornot@slopornot

For Claude Code, Cursor, or Windsurf without the plugin marketplace, use the direct skill install:

npx skills add numen-tech/slopornot

For Codex CLI, Gemini CLI, or OpenCode direct installs, clone the skill into the harness skill directory:

# Codex CLI
mkdir -p ~/.codex/skills
git clone https://github.com/numen-tech/slopornot ~/.codex/skills/agentic-humanizer
 
# Gemini CLI
mkdir -p ~/.gemini/skills
git clone https://github.com/numen-tech/slopornot ~/.gemini/skills/agentic-humanizer
 
# OpenCode
mkdir -p ~/.config/opencode/skills
git clone https://github.com/numen-tech/slopornot ~/.config/opencode/skills/agentic-humanizer

Restart the agent after a direct install so it discovers the skill.

How Do You Connect Slop or Not for Local Checks?

Skip this section if you only want the core humanizer. The Pro tier ships the on-device AI detector tools inside the signed Mac app. Open Settings > Command Line in the app for the current setup command, or call the bundled binary directly:

"/Applications/Slop Or Not.app/Contents/MacOS/slop" status --json

The expected status includes "pro": true and a version number. Recent builds can also include localCachePro. Older builds can expose a legacy entitlement field, but the skill probes Pro by running a real detector call, so both shapes work.

If your agent supports MCP, register the stdio server so the skill can call tools directly. Claude Code:

claude mcp add --transport stdio --scope user SlopOrNot -- "/Applications/Slop Or Not.app/Contents/MacOS/slop" mcp

Codex CLI:

codex mcp add SlopOrNot -- "/Applications/Slop Or Not.app/Contents/MacOS/slop" mcp

Or add this to ~/.codex/config.toml:

[mcp_servers.SlopOrNot]
command = "/Applications/Slop Or Not.app/Contents/MacOS/slop"
args = ["mcp"]

Claude Desktop uses the same command shape in ~/Library/Application Support/Claude/claude_desktop_config.json:

{
  "mcpServers": {
    "SlopOrNot": {
      "command": "/Applications/Slop Or Not.app/Contents/MacOS/slop",
      "args": ["mcp"]
    }
  }
}

After restart, ask the agent to run slop_status. You should see a tool call that reports the local app and Pro state without an error.

How Do You Run the AI Humanizer Loop?

Start the skill, paste the draft, and answer the setup questions. Claude Code plugin installs use:

/slopornot:agentic-humanizer
[paste your draft]

Direct installs and most other clients use:

/agentic-humanizer
[paste your draft]

You can also skip the interview with inline overrides:

/agentic-humanizer dialect=us grade=8 tone=casual length=±10 threshold=20 max=7 [paste]

Common flags:

Flag	What it does
`dialect=us` or `dialect=uk`	Sets American or British English.
`grade=N`	Sets the target Flesch-Kincaid grade.
`tone=casual`, `tone=professional`, or `tone=academic`	Sets the rewrite tone.
`length=±10`, `length=exp`, or `length=trim`	Keeps length close, allows expansion, or allows trimming.
`threshold=N`	Changes the target AI score percentage when Pro is active.
`max=N`	Changes the measured iteration cap when Pro is active.
`voice=/path/to/file.txt`	Uses a writing sample for this run.
`voice=off` or `voice-skip`	Skips voice matching.
`skip-interview`	Uses saved preferences or defaults.

Without Slop or Not Pro, the skill still runs all five rewrite passes and returns the strongest draft; it just avoids claiming on-device AI detector scores it did not measure. With Pro active, the default measured loop runs up to five rewrite iterations. It stops when the AI score is at or below the configured target and the Flesch-Kincaid grade is within the target range, or it returns the best measured pass it found.

Can an AI Humanizer Bypass AI Detectors?

No AI humanizer can guarantee a bypass across GPTZero, Turnitin, Originality.ai, or any other detector. Detectors use different models, thresholds, and input rules. A low score in one tool is not proof that every other tool will agree.

The honest goal is narrower: rewrite the draft for a real reader, remove generic machine-written patterns, and check the result before you use it. Slop or Not adds on-device AI detector feedback to that loop. The agent can see whether the score moved after a pass, then change strategy instead of repeating the same prompt.

That distinction matters. Text Cleanup strips hidden characters and formatting artifacts; it does not rewrite the text. The agentic AI Humanizer skill does the writing work. Slop or Not measures detection, readability, and cleanup locally on your Mac. With Pro active, the final output can show a Text Cleanup summary: invisible characters removed, punctuation normalized, homoglyphs replaced, and British spelling substitutions applied.

Privacy Notes

Slop or Not runs detection, readability, cleanup, and image checks on your Mac. The text sent to the slop CLI or slop mcp server is processed by the local app, not a Slop or Not server.

The rewrite step follows the privacy model of the agent you choose. Claude, Codex, Cursor, Gemini, and other hosted assistants may receive the draft you ask them to revise. If the draft cannot leave your machine, use a local model for rewriting and keep Slop or Not in the same local workflow.

FAQ

Is the agentic AI Humanizer skill free?

Yes. The skill is open source and free to install. The core five-pass rewrite workflow and voice matching do not require Slop or Not. The rewrite step uses the agent subscription or local model you already use. On-device AI detector scoring, readability checks, Text Cleanup, cleanup stats, CLI, and MCP require Slop or Not on Mac with Pro active.

Do I need MCP?

No. MCP is not required for the core humanizer. If you want local AI detector checks with Pro active, MCP is the smoother path because the agent can call detect_text, analyze_readability, and clean_text directly. If your client cannot use MCP, the skill can use the slop CLI, or you can run a manual loop with slop text --json, slop readability --json, and slop cleanup.

Can it match my writing voice?

Yes. Add voice=/path/to/file.txt, save a sample at ~/.agentic-humanizer/voice.txt, or answer the optional voice question when the skill asks. Use at least 50 words. Use 200 or more when you can.

Does Slop or Not upload my draft?

No. Slop or Not checks run locally on your Mac. The rewriting agent has its own data path, so check that agent's privacy model before pasting sensitive text.

Can I run it on Windows or Linux?

The rewrite skill can run anywhere your agent can load it. Voice matching works there too. The on-device AI detector loop needs a Mac with Apple silicon because the CLI and MCP server ship inside the Mac app.

Try It

Install the agentic AI Humanizer skill in your agent, start with /agentic-humanizer or /slopornot:agentic-humanizer, and paste a draft. It works without Slop or Not for core rewriting and voice matching. To add on-device AI detector scores, readability, Text Cleanup, and cleanup stats, install Slop or Not for Mac, activate Pro, and open Settings > Command Line to enable the CLI. For voice matching, read how to humanize AI text in your own voice.