# li-human

Strip the machine fingerprint out of any draft - em dashes, AI slop words, invisible watermark characters - and score it against a five-check detection panel before it goes out. Use whenever text needs to sound human, when the user says humanize, "does this sound like AI", "remove the em dashes", "de-slop this", "will this get flagged", or before any LinkedIn post, comment, reply or DM is shown to the user.

- **Kind:** skill
- **Source:** https://github.com/Jakeschincariol/linkedin-agent-skill
- **Page:** https://forefy.com/skills/d209cbac-600a-4dba-afe3-7555c98fa942
- **API (JSON + files):** https://forefy.com/api/asr/d209cbac-600a-4dba-afe3-7555c98fa942

---

## SKILL.md

---
name: li-human
description: >-
  Strip the machine fingerprint out of any draft - em dashes, AI slop words,
  invisible watermark characters - and score it against a five-check detection
  panel before it goes out. Use whenever text needs to sound human, when the
  user says humanize, "does this sound like AI", "remove the em dashes", "de-slop
  this", "will this get flagged", or before any LinkedIn post, comment, reply
  or DM is shown to the user.
---

# li-human

Two tools live in this folder and they both actually run. Use them. Do not
eyeball this.

```bash
python3 humanize.py draft.txt --report        # clean it, show what changed
python3 detect.py draft.txt                    # score it, five checks
python3 detect.py before.txt after.txt         # prove the delta
```

Both read `slop.json`, which is the lexicon: 100+ stock words and phrases with
plain-English replacements, 17 invisible character classes, 11 typographic
substitutions, and 11 structural tells. It is meant to be edited. If the user
has a word they always use that the lexicon strips, remove it from the file.

## What gets fixed automatically

**1. Invisible characters.** Zero-width spaces and joiners, word joiners,
soft hyphens, byte-order marks, Unicode tag characters, non-breaking and
narrow spaces. A keyboard does not produce these. They survive copy-paste,
they are invisible in every editor, and they are the single most mechanical
thing in generated text. `humanize.py` deletes every one, including any
remaining Unicode format character it does not have a name for.

**2. Typography.** Em dash to comma, en dash to hyphen, curly quotes to
straight, ellipsis to three dots, bullet character to hyphen. The em dash pass
is the one that matters: it collapses ` — ` to `, ` and then cleans up the
double punctuation that leaves behind.

**3. The slop lexicon.** delve, leverage, robust, seamless, crucial, tapestry,
testament to, moreover, "in today's fast-paced world", "let that sink in" and
the rest, each swapped for a plain word, with capitalisation preserved and
URLs left untouched.

## What does NOT get fixed automatically

Structural tells get **flagged, not rewritten**, because changing the shape of
a sentence needs judgement:

- "It's not just X, it's Y" and "not only X but also Y"
- Rule-of-three triads
- Rhetorical one-word question lines: "The result?"
- Rocket, fire, bulb, sparkle and dart emoji
- Hashtag walls
- Reflex engagement bait: "Thoughts?", "Agree?", "Who else?"
- Uniform sentence length and uniform bullet length

That list is your job. Rewrite each flagged line by hand, keeping the meaning,
then re-run `detect.py`. This is the part that moves the score from REVIEW to
PASS, and it is the part a script cannot do.

## The five checks

`detect.py` scores five signals 0-100, higher is more human:

| check | what it measures | machine looks like |
| --- | --- | --- |
| BURSTINESS | sentence-length variation | every sentence the same length |
| SPECIFICITY | numbers, names, concrete markers per 100 words | abstract nouns, no figures |
| SLOP DENSITY | lexicon hits per 100 words | stock vocabulary |
| FINGERPRINT | invisible chars, em dashes, curly quotes per 1k chars | typographically perfect |
| VOICE | contractions, person, structural tells | no contractions, staged reveals |

The verdict weights the mean at 60% and the **weakest single check** at 40%,
because a detector only needs one signal to fire. PASS needs an overall of 70+
with no check below 55.

## Say this honestly

These are five local heuristics modelled on the signals public detectors key
on. They run entirely on the user's machine and nothing is uploaded. They are
**not** GPTZero, Originality, Copyleaks, Winston or Turnitin, they do not call
those APIs, and they cannot promise those verdicts. Fixing what they measure
does tend to move those numbers, because they are measuring the same
underlying things. That is the claim. Do not make a bigger one on the user's
behalf, and do not tell a user their text is undetectable.

## Order of operations

1. `humanize.py draft.txt -o clean.txt --report`
2. Read the structural flags. Rewrite those lines yourself.
3. `detect.py draft.txt clean.txt` to show the before and after.
4. If the verdict is not PASS, fix the weakest check named in the output and
   go again. Two rounds is normal. Five means the draft was written by
   formula, and the fix is a different draft, not more passes.
5. Show the user the cleaned text and the score. Never the score alone.

## detect.py

```python
#!/usr/bin/env python3
"""
detect.py - a five-check panel that scores how machine-written a draft looks.

What this is:  five local heuristics modelled on the signals public AI
detectors actually measure - sentence-length variation, concreteness, stock
vocabulary, typographic fingerprint, and voice. Every score is computed on
your machine from the text alone. Nothing is uploaded.

What this is NOT:  GPTZero, Originality, Copyleaks, Winston or Turnitin.
It does not call their APIs and it cannot promise their verdict. It catches
the things they all key on, which is why fixing them tends to move their
numbers too - but the only honest claim is the one on this line.

Each check returns a HUMAN score from 0 to 100. Higher is better.

Usage
  python3 detect.py draft.txt
  pbpaste | python3 detect.py -
  python3 detect.py draft.txt --json
  python3 detect.py before.txt after.txt      # compare two drafts
"""

import argparse
import json
import os
import re
import statistics
import sys
import unicodedata

HERE = os.path.dirname(os.path.abspath(__file__))
LEX = os.path.join(HERE, "slop.json")

SENT_RE = re.compile(r"[^.!?\n]+[.!?]*")
WORD_RE = re.compile(r"[A-Za-z']+")
CONTRACTIONS = re.compile(r"\b\w+'(?:s|t|re|ve|ll|d|m)\b", re.IGNORECASE)
PRONOUNS = re.compile(r"\b(i|me|my|mine|we|us|our|you|your)\b", re.IGNORECASE)
NUMBERS = re.compile(r"\b\d[\d,.]*%?\b|\$\d")
PROPER = re.compile(r"(?<![.!?]\s)(?<!^)\b[A-Z][a-z]{2,}\b", re.MULTILINE)


def clamp(n):
    return max(0.0, min(100.0, n))


def scale(value, human, machine):
    """Map value onto 0-100 where `human` -> 100 and `machine` -> 0."""
    if human == machine:
        return 50.0
    return clamp((value - machine) / (human - machine) * 100)


def sentences(text):
    return [s.strip() for s in SENT_RE.findall(text) if len(s.split()) > 2]


def words(text):
    return WORD_RE.findall(text)


def check_burstiness(text):
    """Humans vary sentence length hard. Models write even."""
    lens = [len(s.split()) for s in sentences(text)]
    if len(lens) < 4:
        return 50.0, "too short to judge"
    mean = statistics.mean(lens)
    cv = statistics.pstdev(lens) / mean if mean else 0
    score = scale(cv, human=0.70, machine=0.22)
    return score, f"variation {cv:.2f} across {len(lens)} sentences (want 0.55+)"


def check_specificity(text):
    """Numbers, names and concrete nouns. Slop is abstract."""
    w = words(text)
    if len(w) < 25:
        return 50.0, "too short to judge"
    per100 = 100 / len(w)
    hits = len(NUMBERS.findall(text)) + len(set(PROPER.findall(text)))
    density = hits * per100
    score = scale(density, human=6.0, machine=0.5)
    return score, f"{hits} concrete markers, {density:.1f} per 100 words (want 4+)"


def check_slop(text, lex):
    """Stock vocabulary density against the lexicon."""
    w = words(text)
    if not w:
        return 50.0, "empty"
    hits, found = 0, []
    for entry in lex["words"] + lex["phrases"]:
        pattern = re.compile(r"\b" + re.escape(entry["find"]).replace(r"\ ", r"\s+") + r"\b",
                             re.IGNORECASE)
        n = len(pattern.findall(text))
        if n:
            hits += n
            found.append(entry["find"])
    density = hits * 100 / len(w)
    score = scale(density, human=0.0, machine=4.0)
    detail = f"{hits} stock terms, {density:.1f} per 100 words"
    if found:
        detail += " (" + ", ".join(sorted(found)[:4]) + (", ..." if len(found) > 4 else "") + ")"
    return score, detail


def check_fingerprint(text):
    """Characters a phone keyboard does not produce."""
    invisible = sum(1 for c in text if unicodedata.category(c) == "Cf")
    em = text.count("—")
    curly = sum(text.count(c) for c in "‘’“”")
    ellip = text.count("…")
    nbsp = sum(text.count(c) for c in "   ")
    total = invisible * 4 + em * 2 + curly + ellip + nbsp
    per1k = total * 1000 / max(len(text), 1)
    score = scale(per1k, human=0.0, machine=12.0)
    detail = (f"{invisible} invisible, {em} em dash, {curly} curly quote, "
              f"{ellip} ellipsis, {nbsp} hard space")
    return score, detail


def check_voice(text, lex):
    """Contractions, person, and the shapes models default to."""
    w = words(text)
    if len(w) < 25:
        return 50.0, "too short to judge"
    per100 = 100 / len(w)
    contractions = len(CONTRACTIONS.findall(text)) * per100
    person = len(PRONOUNS.findall(text)) * per100
    tells = 0
    names = []
    for s in lex["structures"]:
        try:
            n = len(re.compile(s["regex"], re.MULTILINE).findall(text))
        except re.error:
            continue
        if n:
            tells += n
            names.append(s["id"])
    bullets = [len(b.split()) for b in re.findall(r"(?m)^\s*[-*•]\s+(.+)$", text)]
    uniform = (len(bullets) >= 3 and statistics.pstdev(bullets) < 1.6)
    score = (scale(contractions, human=3.0, machine=0.0) * 0.35
             + scale(person, human=8.0, machine=1.0) * 0.35
             + clamp(100 - tells * 22) * 0.30)
    if uniform:
        score -= 12
        names.append("uniform-bullets")
    detail = (f"{contractions:.1f} contractions, {person:.1f} personal pronouns "
              f"per 100 words, {tells} structural tell(s)")
    if names:
        detail += " [" + ", ".join(names[:4]) + "]"
    return clamp(score), detail


CHECKS = ["BURSTINESS", "SPECIFICITY", "SLOP DENSITY", "FINGERPRINT", "VOICE"]


def run(text, lex):
    results = {}
    results["BURSTINESS"] = check_burstiness(text)
    results["SPECIFICITY"] = check_specificity(text)
    results["SLOP DENSITY"] = check_slop(text, lex)
    results["FINGERPRINT"] = check_fingerprint(text)
    results["VOICE"] = check_voice(text, lex)
    scores = [results[c][0] for c in CHECKS]
    # The weakest check drags the verdict: a detector only needs one signal.
    overall = statistics.mean(scores) * 0.6 + min(scores) * 0.4
    verdict = "PASS" if overall >= 70 and min(scores) >= 55 else (
        "REVIEW" if overall >= 50 else "FLAGGED")
    return results, overall, verdict


def bar(score, width=24):
    filled = round(score / 100 * width)
    return "#" * filled + "." * (width - filled)


def render(results, overall, verdict, label=None, out=sys.stdout):
    title = "AI DETECTION PANEL" + (f"  -  {label}" if label else "")
    print("\n" + title, file=out)
    print("=" * max(len(title), 62), file=out)
    for name in CHECKS:
        score, detail = results[name]
        print(f"  {name:<13} {bar(score)} {score:5.1f}", file=out)
        print(f"  {'':<13} {detail}", file=out)
    print("-" * 62, file=out)
    print(f"  {'HUMAN SCORE':<13} {bar(overall)} {overall:5.1f}   {verdict}", file=out)
    if verdict != "PASS":
        weakest = min(CHECKS, key=lambda c: results[c][0])
        print(f"\n  Weakest signal: {weakest}. Fix that first.", file=out)
    print("", file=out)


def main():
    ap = argparse.ArgumentParser(description="Score how machine-written a draft looks.")
    ap.add_argument("input", nargs="?", default="-", help="file, or - for stdin")
    ap.add_argument("compare", nargs="?", help="second file, to show before/after")
    ap.add_argument("--json", action="store_true")
    ap.add_argument("--lexicon", default=LEX)
    args = ap.parse_args()

    lex = json.load(open(args.lexicon, encoding="utf-8"))
    read = lambda p: sys.stdin.read() if p == "-" else open(p, encoding="utf-8").read()

    targets = [(args.input, read(args.input))]
    if args.compare:
        targets.append((args.compare, read(args.compare)))

    payload = []
    for name, text in targets:
        results, overall, verdict = run(text, lex)
        payload.append({
            "source": name,
            "checks": {k: {"score": round(v[0], 1), "detail": v[1]} for k, v in results.items()},
            "human_score": round(overall, 1),
            "verdict": verdict,
        })

    if args.json:
        print(json.dumps(payload if args.compare else payload[0], indent=2))
        return

    for (name, text), p in zip(targets, payload):
        results, overall, verdict = run(text, lex)
        render(results, overall, verdict, label=os.path.basename(name) if args.compare else None)
    if args.compare:
        a, b = payload
        delta = b["human_score"] - a["human_score"]
        print(f"  {a['human_score']:.1f} {a['verdict']}  ->  "
              f"{b['human_score']:.1f} {b['verdict']}   ({delta:+.1f})\n")

    sys.exit(0 if payload[-1]["verdict"] == "PASS" else 1)


if __name__ == "__main__":
    main()
```

## humanize.py

```python
#!/usr/bin/env python3
"""
humanize.py - strip the machine fingerprint out of a draft.

Three passes, in this order:

  1. INVISIBLE   delete or normalise the characters a human keyboard never
                 produces: zero-width joiners, word joiners, soft hyphens,
                 BOMs, Unicode tag characters, non-breaking and narrow spaces.
                 These survive copy-paste and are the most mechanical tell in
                 any generated text.
  2. TYPOGRAPHIC em dash -> comma, en dash -> hyphen, curly quotes -> straight,
                 ellipsis -> three dots, bullet -> hyphen.
  3. LEXICAL     replace the slop lexicon in slop.json with plain words,
                 preserving capitalisation and leaving URLs untouched.

Structural tells (rule-of-three, "not just X, it's Y", hashtag walls) are
REPORTED, never auto-rewritten - rewriting a sentence's shape needs judgement,
so that is the model's job, not a regex's.

Usage
  python3 humanize.py draft.txt
  python3 humanize.py draft.txt --report
  pbpaste | python3 humanize.py - --report
  python3 humanize.py draft.txt --json
  python3 humanize.py draft.txt -o clean.txt
"""

import argparse
import json
import os
import re
import sys
import unicodedata

HERE = os.path.dirname(os.path.abspath(__file__))
LEX = os.path.join(HERE, "slop.json")

URL_RE = re.compile(r"https?://\S+|www\.\S+|\S+@\S+\.\S+")
SENT_RE = re.compile(r"[^.!?\n]+[.!?]*")


def load_lexicon(path=LEX):
    with open(path, encoding="utf-8") as fh:
        return json.load(fh)


def _cp(spec):
    """'U+200B' -> '\\u200b';  'U+E0000-U+E007F' -> (start, end)."""
    if "-" in spec:
        a, b = spec.split("-")
        return (int(a[2:], 16), int(b[2:], 16))
    return int(spec[2:], 16)


def protect_urls(text):
    """Swap URLs for placeholders so no pass rewrites inside a link."""
    found = []

    def stash(m):
        found.append(m.group(0))
        return f"\x00URL{len(found) - 1}\x00"

    return URL_RE.sub(stash, text), found


def restore_urls(text, found):
    for i, url in enumerate(found):
        text = text.replace(f"\x00URL{i}\x00", url)
    return text


def pass_invisible(text, lex):
    """Delete or space-normalise invisible characters. Returns (text, hits)."""
    hits = []
    for entry in lex["invisible"]:
        cp = _cp(entry["cp"])
        if isinstance(cp, tuple):
            pattern = "[" + re.escape(chr(cp[0])) + "-" + re.escape(chr(cp[1])) + "]"
        else:
            pattern = re.escape(chr(cp))
        n = len(re.findall(pattern, text))
        if n:
            hits.append({"name": entry["cp"] + " " + entry["name"], "count": n,
                         "action": entry["action"]})
            text = re.sub(pattern, "" if entry["action"] == "delete" else " ", text)
    # Any remaining Cf (format) character is invisible by definition.
    stray = [c for c in text if unicodedata.category(c) == "Cf"]
    if stray:
        hits.append({"name": "other invisible format chars", "count": len(stray),
                     "action": "delete"})
        text = "".join(c for c in text if unicodedata.category(c) != "Cf")
    return text, hits


def pass_typographic(text, lex):
    hits = []
    for entry in lex["typographic"]:
        ch = entry["from"]
        n = text.count(ch)
        if not n:
            continue
        hits.append({"name": f"{ch} {entry['name']}", "count": n, "to": entry["to"].strip() or "(space)"})
        if ch == "—":
            # " word — word " and "word—word" both collapse to a comma + space.
            text = re.sub(r"\s*—\s*", ", ", text)
        elif ch == "–":
            text = re.sub(r"\s*–\s*(?=\d)", "-", text)      # 5–10  -> 5-10
            text = re.sub(r"\s+–\s+", ", ", text)            # used as em dash
            text = text.replace("–", "-")
        else:
            text = text.replace(ch, entry["to"])
    # A comma inserted before existing punctuation reads wrong.
    text = re.sub(r",\s*([,.;:!?])", r"\1", text)
    text = re.sub(r",\s*\n", "\n", text)
    return text, hits


def _match_case(src, repl):
    if not repl:
        return repl
    if src.isupper() and len(src) > 1:
        return repl.upper()
    if src[0].isupper():
        return repl[0].upper() + repl[1:]
    return repl


def pass_lexical(text, lex):
    """Replace slop words and phrases. Longest first so phrases win."""
    hits = []
    entries = sorted(lex["phrases"] + lex["words"],
                     key=lambda e: len(e["find"]), reverse=True)
    for entry in entries:
        find = entry["find"]
        pattern = re.compile(r"\b" + re.escape(find).replace(r"\ ", r"\s+") + r"\b",
                             re.IGNORECASE)
        found = pattern.findall(text)
        if not found:
            continue
        hits.append({"find": find, "replace": entry["replace"] or "(deleted)",
                     "count": len(found), "family": entry["family"]})
        text = pattern.sub(lambda m: _match_case(m.group(0), entry["replace"]), text)
    # Clean up after deletions.
    text = re.sub(r"[ \t]{2,}", " ", text)
    text = re.sub(r"(?m)^[ \t]*([,.;:])\s*", "", text)
    text = re.sub(r"\s+([,.;:!?])", r"\1", text)
    text = re.sub(r"(?m)^[ \t]+$", "", text)
    text = re.sub(r"\n{3,}", "\n\n", text)
    # An em dash that became a comma, followed by a sentence connective, leaves
    # a splice ("is important, also, it's proof"). Promote it to a full stop.
    text = re.sub(r",\s*(also|so|still|basically|in the end)\s*,\s*",
                  lambda m: ". " + m.group(1)[0].upper() + m.group(1)[1:] + ", ", text)
    return text, hits


def scan_structures(text, lex):
    flags = []
    for s in lex["structures"]:
        try:
            pattern = re.compile(s["regex"], re.MULTILINE)
        except re.error:
            continue
        found = pattern.findall(text)
        if found:
            flags.append({"name": s["name"], "count": len(found), "fix": s["fix"]})
    # Sentence-length uniformity is structural too.
    lens = [len(s.split()) for s in SENT_RE.findall(text) if len(s.split()) > 2]
    if len(lens) >= 4:
        mean = sum(lens) / len(lens)
        var = sum((n - mean) ** 2 for n in lens) / len(lens)
        cv = (var ** 0.5) / mean if mean else 0
        if cv < 0.35:
            flags.append({
                "name": f"Uniform sentence length (variation {cv:.2f})",
                "count": len(lens),
                "fix": "Break one sentence in half. Let another run long. Machines write even.",
            })
    return flags


def humanize(text, lex):
    text, urls = protect_urls(text)
    text, inv = pass_invisible(text, lex)
    text, typo = pass_typographic(text, lex)
    text, lexi = pass_lexical(text, lex)
    text = restore_urls(text, urls)
    return text.strip() + "\n", {
        "invisible": inv,
        "typographic": typo,
        "lexical": lexi,
        "structures": scan_structures(text, lex),
    }


def render_report(report, out=sys.stderr):
    def head(title):
        print(f"\n{title}\n" + "-" * len(title), file=out)

    total = sum(h["count"] for h in report["invisible"]) \
        + sum(h["count"] for h in report["typographic"]) \
        + sum(h["count"] for h in report["lexical"])

    head("HUMANIZE REPORT")
    print(f"{total} machine artefacts removed, "
          f"{len(report['structures'])} structural tells flagged for rewrite", file=out)

    if report["invisible"]:
        head("1. INVISIBLE CHARACTERS")
        for h in report["invisible"]:
            print(f"  {h['count']:>3}x  {h['name']}  -> {h['action']}", file=out)
    if report["typographic"]:
        head("2. TYPOGRAPHY")
        for h in report["typographic"]:
            print(f"  {h['count']:>3}x  {h['name']}  -> {h['to']}", file=out)
    if report["lexical"]:
        head("3. SLOP LEXICON")
        for h in report["lexical"]:
            print(f"  {h['count']:>3}x  {h['find']}  -> {h['replace']}   [{h['family']}]", file=out)
    if report["structures"]:
        head("4. STRUCTURAL TELLS  (not auto-fixed - rewrite these yourself)")
        for h in report["structures"]:
            print(f"  {h['count']:>3}x  {h['name']}\n        {h['fix']}", file=out)
    if not any(report.values()):
        head("CLEAN")
        print("  Nothing to strip.", file=out)
    print("", file=out)


def main():
    ap = argparse.ArgumentParser(description="Strip the machine fingerprint out of a draft.")
    ap.add_argument("input", nargs="?", default="-", help="file, or - for stdin")
    ap.add_argument("-o", "--out", help="write cleaned text here instead of stdout")
    ap.add_argument("--report", action="store_true", help="print what changed, to stderr")
    ap.add_argument("--json", action="store_true", help="emit {text, report} as JSON")
    ap.add_argument("--lexicon", default=LEX, help="path to slop.json")
    args = ap.parse_args()

    raw = sys.stdin.read() if args.input == "-" else open(args.input, encoding="utf-8").read()
    lex = load_lexicon(args.lexicon)
    clean, report = humanize(raw, lex)

    if args.json:
        print(json.dumps({"text": clean, "report": report}, indent=2, ensure_ascii=False))
        return
    if args.out:
        with open(args.out, "w", encoding="utf-8") as fh:
            fh.write(clean)
        print(f"wrote {args.out}", file=sys.stderr)
    else:
        sys.stdout.write(clean)
    if args.report:
        render_report(report)


if __name__ == "__main__":
    main()
```

## slop.json

```json
{
  "version": "1.0",
  "note": "Lexicon used by humanize.py and detect.py. Replacements are plain-English and keep the sentence grammatical. Edit freely - it is meant to be yours.",
  "invisible": [
    {
      "cp": "U+200B",
      "name": "ZERO WIDTH SPACE",
      "action": "delete"
    },
    {
      "cp": "U+200C",
      "name": "ZERO WIDTH NON-JOINER",
      "action": "delete"
    },
    {
      "cp": "U+200D",
      "name": "ZERO WIDTH JOINER",
      "action": "delete"
    },
    {
      "cp": "U+2060",
      "name": "WORD JOINER",
      "action": "delete"
    },
    {
      "cp": "U+FEFF",
      "name": "ZERO WIDTH NO-BREAK SPACE / BOM",
      "action": "delete"
    },
    {
      "cp": "U+00AD",
      "name": "SOFT HYPHEN",
      "action": "delete"
    },
    {
      "cp": "U+180E",
      "name": "MONGOLIAN VOWEL SEPARATOR",
      "action": "delete"
    },
    {
      "cp": "U+061C",
      "name": "ARABIC LETTER MARK",
      "action": "delete"
    },
    {
      "cp": "U+200E",
      "name": "LEFT-TO-RIGHT MARK",
      "action": "delete"
    },
    {
      "cp": "U+200F",
      "name": "RIGHT-TO-LEFT MARK",
      "action": "delete"
    },
    {
      "cp": "U+E0000-U+E007F",
      "name": "TAG CHARACTERS (invisible ASCII smuggling)",
      "action": "delete"
    },
    {
      "cp": "U+00A0",
      "name": "NO-BREAK SPACE",
      "action": "space"
    },
    {
      "cp": "U+202F",
      "name": "NARROW NO-BREAK SPACE",
      "action": "space"
    },
    {
      "cp": "U+2009",
      "name": "THIN SPACE",
      "action": "space"
    },
    {
      "cp": "U+2007",
      "name": "FIGURE SPACE",
      "action": "space"
    },
    {
      "cp": "U+2003",
      "name": "EM SPACE",
      "action": "space"
    },
    {
      "cp": "U+2002",
      "name": "EN SPACE",
      "action": "space"
    }
  ],
  "typographic": [
    {
      "from": "—",
      "name": "EM DASH",
      "to": ", "
    },
    {
      "from": "–",
      "name": "EN DASH",
      "to": "-"
    },
    {
      "from": "‘",
      "name": "LEFT SINGLE QUOTE",
      "to": "'"
    },
    {
      "from": "’",
      "name": "RIGHT SINGLE QUOTE",
      "to": "'"
    },
    {
      "from": "“",
      "name": "LEFT DOUBLE QUOTE",
      "to": "\""
    },
    {
      "from": "”",
      "name": "RIGHT DOUBLE QUOTE",
      "to": "\""
    },
    {
      "from": "…",
      "name": "HORIZONTAL ELLIPSIS",
      "to": "..."
    },
    {
      "from": "′",
      "name": "PRIME",
      "to": "'"
    },
    {
      "from": "·",
      "name": "MIDDLE DOT",
      "to": "-"
    },
    {
      "from": "•",
      "name": "BULLET",
      "to": "-"
    },
    {
      "from": "→",
      "name": "RIGHTWARDS ARROW",
      "to": "->"
    }
  ],
  "words": [
    {
      "find": "delve into",
      "replace": "look at",
      "family": "verbs"
    },
    {
      "find": "delve",
      "replace": "look",
      "family": "verbs"
    },
    {
      "find": "dive deep into",
      "replace": "get into",
      "family": "verbs"
    },
    {
      "find": "deep dive",
      "replace": "breakdown",
      "family": "nouns"
    },
    {
      "find": "leverage",
      "replace": "use",
      "family": "verbs"
    },
    {
      "find": "leveraging",
      "replace": "using",
      "family": "verbs"
    },
    {
      "find": "utilize",
      "replace": "use",
      "family": "verbs"
    },
    {
      "find": "utilizing",
      "replace": "using",
      "family": "verbs"
    },
    {
      "find": "harness",
      "replace": "use",
      "family": "verbs"
    },
    {
      "find": "embark on",
      "replace": "start",
      "family": "verbs"
    },
    {
      "find": "foster",
      "replace": "build",
      "family": "verbs"
    },
    {
      "find": "facilitate",
      "replace": "help",
      "family": "verbs"
    },
    {
      "find": "showcase",
      "replace": "show",
      "family": "verbs"
    },
    {
      "find": "unlock",
      "replace": "get",
      "family": "verbs"
    },
    {
      "find": "elevate",
      "replace": "improve",
      "family": "verbs"
    },
    {
      "find": "streamline",
      "replace": "simplify",
      "family": "verbs"
    },
    {
      "find": "spearhead",
      "replace": "lead",
      "family": "verbs"
    },
    {
      "find": "underscore",
      "replace": "show",
      "family": "verbs"
    },
    {
      "find": "underscores",
      "replace": "shows",
      "family": "verbs"
    },
    {
      "find": "cultivate",
      "replace": "build",
      "family": "verbs"
    },
    {
      "find": "amplify",
      "replace": "boost",
      "family": "verbs"
    },
    {
      "find": "curate",
      "replace": "pick",
      "family": "verbs"
    },
    {
      "find": "curated",
      "replace": "picked",
      "family": "verbs"
    },
    {
      "find": "empower",
      "replace": "let",
      "family": "verbs"
    },
    {
      "find": "revolutionize",
      "replace": "change",
      "family": "verbs"
    },
    {
      "find": "transformative",
      "replace": "big",
      "family": "adjectives"
    },
    {
      "find": "robust",
      "replace": "solid",
      "family": "adjectives"
    },
    {
      "find": "seamless",
      "replace": "clean",
      "family": "adjectives"
    },
    {
      "find": "seamlessly",
      "replace": "cleanly",
      "family": "adjectives"
    },
    {
      "find": "pivotal",
      "replace": "key",
      "family": "adjectives"
    },
    {
      "find": "crucial",
      "replace": "important",
      "family": "adjectives"
    },
    {
      "find": "vital",
      "replace": "important",
      "family": "adjectives"
    },
    {
      "find": "comprehensive",
      "replace": "complete",
      "family": "adjectives"
    },
    {
      "find": "cutting-edge",
      "replace": "new",
      "family": "adjectives"
    },
    {
      "find": "game-changing",
      "replace": "big",
      "family": "adjectives"
    },
    {
      "find": "game-changer",
      "replace": "big deal",
      "family": "nouns"
    },
    {
      "find": "groundbreaking",
      "replace": "new",
      "family": "adjectives"
    },
    {
      "find": "unparalleled",
      "replace": "unmatched",
      "family": "adjectives"
    },
    {
      "find": "invaluable",
      "replace": "useful",
      "family": "adjectives"
    },
    {
      "find": "meticulous",
      "replace": "careful",
      "family": "adjectives"
    },
    {
      "find": "meticulously",
      "replace": "carefully",
      "family": "adjectives"
    },
    {
      "find": "myriad of",
      "replace": "many",
      "family": "adjectives"
    },
    {
      "find": "myriad",
      "replace": "many",
      "family": "adjectives"
    },
    {
      "find": "multifaceted",
      "replace": "complicated",
      "family": "adjectives"
    },
    {
      "find": "holistic",
      "replace": "whole",
      "family": "adjectives"
    },
    {
      "find": "bespoke",
      "replace": "custom",
      "family": "adjectives"
    },
    {
      "find": "innovative",
      "replace": "new",
      "family": "adjectives"
    },
    {
      "find": "profound",
      "replace": "big",
      "family": "adjectives"
    },
    {
      "find": "remarkable",
      "replace": "notable",
      "family": "adjectives"
    },
    {
      "find": "compelling",
      "replace": "convincing",
      "family": "adjectives"
    },
    {
      "find": "tapestry",
      "replace": "mix",
      "family": "nouns"
    },
    {
      "find": "realm",
      "replace": "world",
      "family": "nouns"
    },
    {
      "find": "landscape",
      "replace": "market",
      "family": "nouns"
    },
    {
      "find": "journey",
      "replace": "process",
      "family": "nouns"
    },
    {
      "find": "testament to",
      "replace": "proof of",
      "family": "nouns"
    },
    {
      "find": "cornerstone",
      "replace": "base",
      "family": "nouns"
    },
    {
      "find": "beacon",
      "replace": "signal",
      "family": "nouns"
    },
    {
      "find": "paradigm",
      "replace": "model",
      "family": "nouns"
    },
    {
      "find": "synergy",
      "replace": "overlap",
      "family": "nouns"
    },
    {
      "find": "ecosystem",
      "replace": "system",
      "family": "nouns"
    },
    {
      "find": "treasure trove",
      "replace": "pile",
      "family": "nouns"
    },
    {
      "find": "plethora of",
      "replace": "lots of",
      "family": "nouns"
    },
    {
      "find": "plethora",
      "replace": "lots",
      "family": "nouns"
    },
    {
      "find": "arsenal",
      "replace": "set",
      "family": "nouns"
    },
    {
      "find": "north star",
      "replace": "goal",
      "family": "nouns"
    },
    {
      "find": "moreover",
      "replace": "also",
      "family": "connectives"
    },
    {
      "find": "furthermore",
      "replace": "also",
      "family": "connectives"
    },
    {
      "find": "additionally",
      "replace": "also",
      "family": "connectives"
    },
    {
      "find": "nevertheless",
      "replace": "still",
      "family": "connectives"
    },
    {
      "find": "consequently",
      "replace": "so",
      "family": "connectives"
    },
    {
      "find": "thus",
      "replace": "so",
      "family": "connectives"
    },
    {
      "find": "hence",
      "replace": "so",
      "family": "connectives"
    },
    {
      "find": "notably",
      "replace": "",
      "family": "connectives"
    },
    {
      "find": "importantly",
      "replace": "",
      "family": "connectives"
    },
    {
      "find": "ultimately",
      "replace": "in the end",
      "family": "connectives"
    },
    {
      "find": "in essence",
      "replace": "basically",
      "family": "connectives"
    },
    {
      "find": "in conclusion",
      "replace": "so",
      "family": "connectives"
    },
    {
      "find": "when it comes to",
      "replace": "with",
      "family": "connectives"
    },
    {
      "find": "in order to",
      "replace": "to",
      "family": "connectives"
    },
    {
      "find": "a wide range of",
      "replace": "many",
      "family": "connectives"
    },
    {
      "find": "it is worth noting that",
      "replace": "note that",
      "family": "connectives"
    }
  ],
  "phrases": [
    {
      "find": "in today's fast-paced world",
      "replace": "right now",
      "family": "openers"
    },
    {
      "find": "in today's digital age",
      "replace": "right now",
      "family": "openers"
    },
    {
      "find": "in the ever-evolving landscape of",
      "replace": "in",
      "family": "openers"
    },
    {
      "find": "in the ever-changing world of",
      "replace": "in",
      "family": "openers"
    },
    {
      "find": "let's face it",
      "replace": "",
      "family": "openers"
    },
    {
      "find": "here's the thing",
      "replace": "",
      "family": "openers"
    },
    {
      "find": "picture this",
      "replace": "",
      "family": "openers"
    },
    {
      "find": "imagine a world where",
      "replace": "imagine if",
      "family": "openers"
    },
    {
      "find": "it's no secret that",
      "replace": "",
      "family": "openers"
    },
    {
      "find": "we've all been there",
      "replace": "",
      "family": "openers"
    },
    {
      "find": "let that sink in",
      "replace": "",
      "family": "closers"
    },
    {
      "find": "the choice is yours",
      "replace": "",
      "family": "closers"
    },
    {
      "find": "food for thought",
      "replace": "",
      "family": "closers"
    },
    {
      "find": "the future is here",
      "replace": "",
      "family": "closers"
    },
    {
      "find": "what are your thoughts?",
      "replace": "",
      "family": "closers"
    },
    {
      "find": "drop a comment below",
      "replace": "",
      "family": "closers"
    },
    {
      "find": "at the end of the day",
      "replace": "",
      "family": "closers"
    },
    {
      "find": "needle-moving",
      "replace": "useful",
      "family": "closers"
    },
    {
      "find": "move the needle",
      "replace": "make a difference",
      "family": "closers"
    },
    {
      "find": "a testament to",
      "replace": "proof of",
      "family": "nouns"
    },
    {
      "find": "had the opportunity to",
      "replace": "got to",
      "family": "connectives"
    },
    {
      "find": "i am excited to announce",
      "replace": "",
      "family": "openers"
    },
    {
      "find": "i'm thrilled to share",
      "replace": "",
      "family": "openers"
    },
    {
      "find": "i am humbled to",
      "replace": "",
      "family": "openers"
    },
    {
      "find": "without further ado",
      "replace": "",
      "family": "openers"
    },
    {
      "find": "buckle up",
      "replace": "",
      "family": "openers"
    },
    {
      "find": "the harsh truth is",
      "replace": "",
      "family": "openers"
    },
    {
      "find": "spoiler alert",
      "replace": "",
      "family": "openers"
    },
    {
      "find": "read that again",
      "replace": "",
      "family": "closers"
    },
    {
      "find": "you're welcome",
      "replace": "",
      "family": "closers"
    },
    {
      "find": "save this for later",
      "replace": "",
      "family": "closers"
    },
    {
      "find": "who's with me",
      "replace": "",
      "family": "closers"
    }
  ],
  "structures": [
    {
      "id": "not-just",
      "regex": "(?i)\\bit'?s not (?:just|only) [^.!?\\n]{2,60}[,.] it'?s\\b",
      "name": "\"It's not just X, it's Y\"",
      "fix": "Say the one thing you actually mean. Delete the setup half."
    },
    {
      "id": "not-only",
      "regex": "(?i)\\bnot only\\b[^.!?\\n]{2,80}\\bbut also\\b",
      "name": "\"Not only X but also Y\"",
      "fix": "Split into two sentences, or drop the weaker half."
    },
    {
      "id": "isnt-about",
      "regex": "(?i)\\b(?:this|it) is(?:n'?t| not) about [^.!?\\n]{2,60}[.!] it'?s about\\b",
      "name": "\"It isn't about X. It's about Y.\"",
      "fix": "Lead with Y. The reversal is the tell."
    },
    {
      "id": "the-result",
      "regex": "(?i)^\\s*(?:the result|the best part|the kicker|the catch)\\?\\s*$",
      "name": "One-word rhetorical question line",
      "fix": "Answer it inline instead of staging a reveal."
    },
    {
      "id": "heres-what",
      "regex": "(?i)\\bhere'?s what (?:i|we) learned\\b",
      "name": "\"Here's what I learned\"",
      "fix": "State the lesson. Skip the announcement."
    },
    {
      "id": "triad",
      "regex": "(?i)\\b\\w+, \\w+, and \\w+\\.",
      "name": "Rule-of-three triad",
      "fix": "Two items or four. Three is the model's default rhythm."
    },
    {
      "id": "rocket",
      "regex": "[\\U0001F680\\U0001F525\\U0001F4A1\\u2728\\U0001F3AF]",
      "name": "Rocket / fire / bulb / sparkle / dart emoji",
      "fix": "One emoji max, and not one of these five."
    },
    {
      "id": "hashtag-wall",
      "regex": "(?:#\\w+[ \\t]*){5,}",
      "name": "Hashtag wall (5+)",
      "fix": "Three hashtags. LinkedIn stopped rewarding the wall."
    },
    {
      "id": "as-an-ai",
      "regex": "(?i)\\bas an ai\\b|\\bas a language model\\b",
      "name": "Model self-reference",
      "fix": "Delete the sentence."
    },
    {
      "id": "in-summary",
      "regex": "(?i)^\\s*(?:in summary|to summarize|in conclusion)\\b",
      "name": "Essay-closer opener",
      "fix": "A post does not need a conclusion paragraph."
    },
    {
      "id": "engagement-bait",
      "regex": "(?i)\\b(?:agree\\?|thoughts\\?|who else\\b|am i wrong\\?)\\s*$",
      "name": "Reflex engagement bait",
      "fix": "Ask a question only you could ask, about the specific thing above it."
    }
  ]
}
```