agentleFS
abduallahJamalabduallahJamal/humanizer-skills/skills/ai-writing-detector/SKILL.md

ai-writing-detector

Skill0 starsChanged 10 days agoMITView on GitHub

Audit and score English or Arabic text for AI-writing patterns without rewriting it. Runs a 12-metric stylometric analysis plus a 9-signal forensic rubric with quoted evidence and sentence-level heatmap. Use when asked if text sounds like AI, why it gets flagged, or to audit drafts in detect-only mode (فحص وكشف بصمات الذكاء الاصطناعي في النصوص العربية والإنجليزية بدون تعديل النص).

6 other projects use a version of this.

Adopted by

Most recently taken up by

The file

We split this skill into the rules it gives an agent

1 rule. 0 are common, 0 are distinctive, and 1 is very specific to this project.

  • 0

    Common

    Most of the 4 edited versions we read keep these.

  • 0

    Distinctive

    This author's own choices. Worth a look.

  • 1

    Very specific

    Tied to their project's names, files or tools. Probably skip.

  • Common
  • Distinctive
  • Very specific
  1. Very specific:

    Read references/scoring-rubric.md and score each of the 9 categories from 0 (human-consistent) to 3 (strong AI pattern):

    line 31

“Common” is measured: kept by at least 6 in 10 of the 4 edited versions of this file we could read. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.

Keep what fits and drop the rest. Your version, ready for your agents.
Files
SKILL.md

AI Writing Detector — Forensic Audit & Stylometric Scorer (EN & AR)

Analyze text for statistical and rhetorical signs of AI generation without rewriting or altering the source text. Every flagged signal must cite exact quotes or measured counts.

Detect-Only Contract

  1. Never rewrite the text unless the user explicitly asks for a rewrite after seeing the report (in which case hand off to humanize-text or academic-humanizer).
  2. Signals, not courtroom proof. Commercial and open-source AI detectors exhibit high false-positive rates on non-native writers, deadline-pressed authors, and rigid technical templates. Present findings as stylistic & statistical signals that explain why readers or detectors flag a passage — never as an accusation of misconduct.
  3. Separate real tells from judgment calls. Distinguish Strong Markers (chat residue, Tier-1A clusters, negative parallelisms, uniform 15-word pulse, Arabic تم + مصدر / لا يقتصر على calques) from Context-Dependent Judgment Calls (legitimate technical terms like robust in statistics, numbered API steps, or house-style em dashes).

Workflow

Step 1: Run Deterministic 12-Metric Stylometric Scan

Whenever python3 is available, run the bundled local analyzer first so your numbers (word counts, sentence lengths, burstiness CV, bigram diversity, punctuation counts, and flagged sentences) are exact rather than estimated:

python3 <skill-dir>/scripts/detect_ai_metrics.py --file input.txt --format markdown

(Or pipe stdin: cat input.txt | python3 <skill-dir>/scripts/detect_ai_metrics.py)

Step 2: Score the 9 Forensic Signals (0–27 Rubric)

Read references/scoring-rubric.md and score each of the 9 categories from 0 (human-consistent) to 3 (strong AI pattern):

  • Signal A: Lexical Predictability & Tier-1 AI Vocabulary (EN & AR)
  • Signal B: Burstiness Deficit & Metronomic Sentence Lengths
  • Signal C: Institutional Hedge Density
  • Signal D: Imposed Document Architecture & Forced Triads
  • Signal E: Specificity & Concrete-Anchor Deficit
  • Signal F: Mechanical Transition & Announcement Fingerprint (Furthermore, Turns out, علاوة على ذلك, تجدر الإشارة)
  • Signal G: Punctuation & Typography Fingerprint (Em dashes, semicolons, bold-colon lists)
  • Signal H: Register Collapse & Translation Calques (e.g. Arabic تم + مصدر, يقوم بـ, بشكل + صفة, synonym cycling)
  • Signal I: RLHF "Helpful Assistant" & Rhetorical Scaffolding (Not X, but Y, mini-aphorism closers, both-sides diplomacy, chat leftovers)

Double-counting rule: A single phrase may fire at most two distinct categories when genuinely diagnostic for both, and never counts twice inside the same category.

Step 3: Deliver the Structured Forensic Report

Format your response in the user's preferred language (Arabic if they asked in Arabic, English otherwise) with four concise sections:

  1. Overall Signal Summary:
    • Forensic Score: X / 27 (0–5 Low / Human-consistent, 6–11 Mild / Mixed, 12–18 Moderate-High AI Footprint, 19–27 Heavy Unedited LLM Output)
    • Local Stylometric Human Score from detect_ai_metrics.py (0–100%) + sentence length sequence.
  2. Signal Breakdown Table (A–I): Score (0–3) + exact quoted evidence from the text (or Clean if 0).
  3. Top Flagged Sentences (Heatmap): The 3–5 highest-risk sentences and the exact fix principle for each (without rewriting the whole piece).
  4. Protected / Intentional Exceptions: Note any matches that are actually fine in context (e.g., domain terms, legal/academic conventions) so the writer does not over-edit good prose.

Download now (ZIP)

Does it work?

Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.

No reviews yet

People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.

  • It worked
  • Partly
  • It didn't
What did you use it for? e.g. Reviewing pull requests in a Rails app

Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.

Your trail

Files you open will collect here, so you can find them again.

Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.

Now do this for your own.

Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.