li-human

Before you start: read this whole install prompt, tell me in a few lines what this skill does and what installing it involves, and wait for my go-ahead before you do anything else. Prerequisites: none for the install itself: no account, no API key, no paid service, and nothing else to install first. The skill is free under the MIT licence. For using the skill afterwards you need Python 3 only: both bundled scripts (humanize.py, detect.py) use the Python standard library only, with no packages to install and no network access. There is no LinkedIn account, login, API key or external detector service involved at any point. Install "li-human" for me. It is the humanizer in the LinkedIn agent skill pack: it strips the machine fingerprint out of a draft before the draft goes out. Two Python tools actually run, both reading the bundled lexicon slop.json (113 stock words and phrases with plain-English replacements: 81 in "words" and 32 in "phrases", plus 17 invisible character classes, 11 typographic substitutions and 11 structural-tell patterns). humanize.py cleans a draft in three passes, in order: (1) invisible characters deleted or normalised (zero-width spaces and joiners, word joiners, soft hyphens, byte-order marks, Unicode tag characters, non-breaking and narrow spaces, plus any remaining Unicode format character); (2) typography normalised (em dash to comma, en dash to hyphen, curly quotes to straight, ellipsis to three dots, bullet to hyphen); (3) the slop lexicon replaced with plain words, capitalisation preserved and URLs left untouched. Structural tells ("It's not just X, it's Y", rule-of-three triads, one-word rhetorical question lines, rocket/fire/bulb/sparkle/dart emoji, hashtag walls, reflex engagement bait, uniform sentence length) are flagged for a manual rewrite, never auto-rewritten. detect.py then scores the draft on a five-check panel, each check 0-100, higher is more human: BURSTINESS (sentence-length variation), SPECIFICITY (numbers, names and concrete markers per 100 words), SLOP DENSITY (lexicon hits per 100 words), FINGERPRINT (invisible characters, em dashes and curly quotes per 1,000 characters) and VOICE (contractions, personal pronouns, structural tells). The overall Human Score is the mean of the five weighted at 60% plus the weakest single check at 40%; PASS needs an overall of 70 or more with no check below 55, otherwise the verdict is REVIEW or FLAGGED. SKILL.md is explicit, and so is this prompt: these are five local heuristics modelled on the signals public detectors key on, they are not GPTZero, Originality, Copyleaks, Winston or Turnitin, they do not call those APIs, and they cannot promise those verdicts or that a text is undetectable. Repository: https://github.com/avi691/linkedin-agent-skill (the skill is the whole skills/li-human/ folder at the pinned commit: SKILL.md plus humanize.py, detect.py and slop.json, 4 files) Pinned version: install the code at commit e4ec4fb6f1b0e227eb9b14b31b735072db5df9cb exactly. Do not substitute a newer version. If this commit is unreachable, stop and tell me. 1. Clone https://github.com/avi691/linkedin-agent-skill into a temporary folder, check out commit e4ec4fb6f1b0e227eb9b14b31b735072db5df9cb, and copy the skills/li-human/ folder from that checkout. Then compare every file in the folder with the fingerprints recorded for this listing on 2026-10-19: [FINGERPRINTS] - SKILL.md: SHA-256 2ebda9ff1b63f0d567cd8e30b6fc1465e05c46e166809ea0569cba3591d73d90 - detect.py: SHA-256 d8cf740741968f25ce49ec6cc65538f53e97f28f6f2ce21b7f21864475fe2075 - humanize.py: SHA-256 8627c813e461459604c5019d28052fbde11a544af86a1c4652b1aff3b38d8757 - slop.json: SHA-256 2d227835a80ca20259c8d255501cb752cdc4232468c12bb9223fccf5acc724bb If a file does not match, it changed since the listing was prepared: do not install it yet, tell me what you found and let me decide. 2. Read SKILL.md in full, read both scripts in full (humanize.py and detect.py) and read slop.json, and tell me in plain words: what the skill does when used, which services it connects to, whether it reads or sends any personal information or keys, and anything that seems unusual. Stop on any red flag and tell me before going further. Confirm these points from the files: - The skill folder holds 4 files only: SKILL.md (4,522 bytes), humanize.py (9,460 bytes), detect.py (8,526 bytes) and slop.json (15,874 bytes). The skill is self-contained in its own folder: both scripts locate slop.json next to themselves (the folder they sit in) and read nothing else from the repository. SKILL.md and the scripts reference no other skill in the pack, not the other 10 li-* skills, not templates/voice.md, not the README and not the root of the repository. (Other skills in the pack send their drafts through /li-human; that is an inbound use of this skill, not a dependency of this skill.) - Nothing is sent anywhere during installation itself; the only download at install time is the skill folder from GitHub at the pinned commit. No script is run during installation, no key is asked for, and there is no telemetry in the folder. - What happens at use time, disclosed here before it ever happens, and using the skill is a separate step that only starts when I ask you to humanize or score a draft: everything runs locally on this machine and nothing leaves this machine. Neither script imports any networking, subprocess or package code: humanize.py imports only argparse, json, os, re, sys and unicodedata; detect.py imports only argparse, json, os, re, statistics, sys and unicodedata, all Python standard library. There is no URL called, no API called and no file uploaded anywhere in either script; the only URL-shaped code is a regular expression in humanize.py that protects URLs already present in my draft text so the cleaning passes do not rewrite inside a link. humanize.py reads exactly two things: the draft I name (a file path argument, or standard input when the argument is "-") and slop.json beside the script (another lexicon path only if I pass --lexicon myself). It writes the cleaned text to standard output, or to the single output file I name with -o/--out; with --report it also prints a change report to standard error, and with --json it prints the cleaned text and the report as JSON to standard output. detect.py reads the one draft I name (or standard input with "-"), plus a second draft file only when I give one for a before/after comparison, and slop.json beside the script; it writes nothing to disk, it only prints the panel (or JSON with --json) to standard output, and it exits with code 0 when the last verdict is PASS and 1 otherwise, so a non-PASS score is a normal result, not a crash. Both scripts compile the structural-tell regular expressions stored in slop.json, so slop.json is a trusted local data file: it is meant to be edited by me (for example to remove a word I always use), and I should only run the scripts with a slop.json I trust. - The order of operations in SKILL.md: run humanize.py with -o and --report, rewrite the flagged structural lines by hand, run detect.py on the before and after files to show the delta, and if the verdict is not PASS, fix the weakest check named in the output and go again (two rounds is normal); show me the cleaned text and the score together, never the score alone. The skill posts nothing itself: it prepares text, and posting to LinkedIn stays a separate manual step for me. - Do not run humanize.py, detect.py or any other script during this install, do not humanize or score any text, and do not install Python or anything else: setup is my decision, later, when I actually ask for a draft to be cleaned. 3. When I say "yes, install", install it as a skill: copy the whole li-human folder into the agent's skills directory, in a folder named "li-human", keeping all 4 files (SKILL.md, humanize.py, detect.py and slop.json together in the same folder, because the scripts find slop.json next to themselves). A manual clone followed by copying the skills/li-human folder into ~/.claude/skills/ is the route described here; any other route is fine only as long as every installed file matches its fingerprint. 4. Verify with no network calls: frontmatter valid (name: li-human), all 4 files in place with matching fingerprints. Do not run humanize.py or detect.py to verify. 5. Report what was installed, where, and what I still need to do myself (for example: nothing for the install itself; to use the skill later, have Python 3 available, give the skill a draft file or paste the draft as standard input, optionally edit slop.json to make the lexicon mine, rewrite any flagged structural lines myself with the agent's help, and post the result to LinkedIn myself; using the skill is a separate step that only starts when I ask for it, it runs entirely locally, and the use-time behaviour disclosed above, local reads of my draft and slop.json and a local write only to an output file I name, is all that ever happens). GitHub is optional: if I have a GitHub account or the gh CLI, you may use it; otherwise public access is fine. Never require it unless it's in the prerequisites above. Rules: don't touch anything outside the temp folder and the install target. If anything looks off, stop and ask me.