Human

human

Two tools live in this folder and they both actually run. Use them. Do not eyeball this. Paths below are relative to this skill's base directory, which is given to you when the skill is invoked.

python3 humanize.py draft.txt --report        # clean it, show what changed
python3 detect.py draft.txt                    # score it, six checks
python3 detect.py before.txt after.txt         # prove the delta
pbpaste | python3 humanize.py - --report       # nothing touches disk

Both read slop.json, which is the lexicon: 228 stock words and phrases with plain-English replacements, 17 invisible character classes, 11 typographic substitutions, and 11 structural tells.

What this is for at Syncx

Anything a customer, a teammate or the public reads: support replies about a feed or profile failure, investigation write-ups, release notes, help-centre articles, PR and commit bodies, and marketing copy. Run it last, after the content is correct. A reply that sounds human but names the wrong profile is still a bad reply.

Pair it with support-reply for customer answers: that skill decides what to say, this one decides how it reads.

Do not run it on these

• Code, config, SQL, logs, stack traces. The lexical pass rewrites plain words wherever it finds them, and it does not know it is inside a fence.
• Markdown with inline code spans. It protects URLs, nothing else. A documented character inside backticks gets substituted like prose and the meaning dies. Split the prose out, clean that, put it back.
• Exact error strings and vendor field names. Those are evidence. Keep them byte-for-byte.
• Anything carrying customer data. No API keys, tokens, store domains, account IDs or merchant names in a draft file. Pipe stdin instead, so nothing lands on disk. If you must write a file, mask the identifiers before it exists, not after.

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 the spaced em dash to a comma 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:

• The set-up-then-correct family: "It's not just X, it's Y", "not only X but also Y", "It isn't about X. It's about Y.", "Not because X. But because Y.", "Not by doing X, but by doing Y.", "No X. No Y. Just Z."
• Rule-of-three triads
• Rhetorical lines standing in for a claim: "The result?", "And the fix? Y."
• Model self-reference: "As an AI", "as a language model"
• Essay closers: a paragraph opening with "In summary" or "In conclusion"
• Three or more bullets of near-identical 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 six checks

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

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.

VOICE and STRUCTURE were one check until they were split apart. VOICE now reads prose only, and STRUCTURE reads shape, which works on any text type. A document of tables and command blocks can score 100 on STRUCTURE and near zero on VOICE, and both numbers are telling the truth about different things.

Technical writing fails VOICE first, almost every time. Profile numbers and vendor names carry SPECIFICITY for free, so the score you will be fixing is the one measuring contractions and second person. Write "you'll see" instead of "the user will observe" and most of the gap closes.

Short drafts cannot pass. Under four sentences, BURSTINESS, SPECIFICITY and VOICE all return too short to judge and score a flat 50, which sits below the 55 floor. STRUCTURE has no length guard and will usually read 100 on a short draft, which lifts the mean but cannot lift the floor. A short reply cannot reach PASS no matter how well written it is. On anything short, read FINGERPRINT and STRUCTURE, and ignore the verdict.

Say this honestly

These are six 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.