You can spot AI-written text in three seconds: the 60-word sentence that never lands, "leverage the robust tapestry," the "furthermore / moreover / additionally" drumbeat, the hedge pile ("might arguably perhaps be"). Grammar checkers catch none of this — the grammar is perfect. The register is the problem.
plain-speak is a tiny readability linter that scores text on plainness and flags the offending lines, so you (or your agent) can de-slop a draft before it ships.
git clone https://github.com/hahahahahahahahah6/plain-speak
cd plain-speak
python3 plain_speak.py check draft.txt
$ python3 plain_speak.py check draft.txt
plainness: 45/100 (slop)
sentences: 3 words: 80 reading ease: 30.7
long sentences: 2 buzzwords: 24 hedges: 2
findings:
LINE RULE DETAIL
---- ---- ------
1 long-sentence Sentence has 32 words (limit 25)
1 buzzword AI-tell buzzword: "In today's fast-paced"
1 buzzword AI-tell buzzword: "leverage"
2 buzzword AI-tell buzzword: "tapestry"
2 hedge Hedge word: "arguably"
Four checks: sentences over 25 words, ~45 built-in AI-tell buzzwords (delve, robust, furthermore, in today's fast-paced, …), a Flesch-style reading-ease score folded into the total, and hedge density. --json and --fail-under 80 make it a CI gate for generated docs.
It's also built for agents: the repo ships a SKILL.md agent skill with a reading-level ladder (ELI5 / engineer / one-liner). The CLI detects; the skill rewrites. Loop them — check, rewrite, check again — until the score passes.
One file, stdlib only, Python 3.9+ (grab it via curl if you don't want the clone). 13/13 tests pass. MIT licensed.
Repo: https://github.com/hahahahahahahahah6/plain-speak
What's your favorite AI-tell word — the one that instantly gives a draft away?
Top comments (0)