llmism

Detector and remediator for LLMisms — idioms and tics overused by language models that make prose sound machine-generated: delve, "it's not X, it's Y", em-dash overuse, bolded bullet lead-ins, etc.

Install (local prototype)

pip install -e ".[dev]"

Detect

llmism detect README.md          # console table, exit 1 if findings
llmism detect doc.tex            # .tex → LaTeX, .md → Markdown, else text
llmism detect - --format markdown < draft.txt
llmism detect post.md --json     # machine-readable findings

LINE  CATEGORY   RULE                     MATCH
-----  ---------  -----------------------  ------------------------------------
3      lexical    delve                    'delve'
3      lexical    leverage-verb            'leverage'
5      phrasal    its-not-x-its-y          "It's not a bug, it's "

Fix

llmism fix post.md                    # rewritten text on stdout
llmism fix post.md --in-place         # rewrite the file
llmism fix post.md --json             # {text, fixed, remaining, ...}

• Lexical/phrasal findings with a safe replacement are rewritten deterministically (capitalisation preserved, offsets applied right-to-left).
• Structural findings (em-dash density, bold-lead-in bullets, rhetorical Q&A pairs, low burstiness) have no safe deterministic rewrite. They are reported as remaining unless you pass --llm:

pip install -e ".[llm]"   # optional anthropic extra
export ANTHROPIC_API_KEY=...
llmism fix post.md --llm --in-place

What is detected

Tier	Rules
lexical	delve, tapestry, pivotal, leverage, seamless, boundaries, robust, load-bearing, ...
phrasal	"it's not X, it's Y", "not only X but also Y", -ing tail clauses, vague attributions ("studies suggest"), cataloguing lead-ins ("uses several mechanisms... These methods..."), empty pivots ("it's worth noting"), significance inflation ("stands as a testament to"), Furthermore/Moreover, ...
structural	em-dash overuse (per paragraph), transition-word clusters, sentence-opener repetition, hedge stacking, ≥3 bullets with **bold** lead-ins, self-answered rhetorical questions, uniform sentence rhythm (low burstiness), uniform paragraph sizes, mechanical short/long cadence, synonym cycling, degenerate repetition

Pattern data lives in llmism/data/patterns.yaml — extend the list without touching code. Code fences (Markdown) and math environments, verbatim blocks and % comments (LaTeX) are never scanned or rewritten. Several structural rules are inspired by the sloptrim pattern catalogue.

Library use

from llmism import Detector, Remediator

findings = Detector().scan(text, "markdown")
result = Remediator().fix(text, findings)
print(result.remaining)   # findings left for manual fixing

Development

pytest -q --cov=llmism
ruff check . && mypy llmism

MIT licensed. See LICENSE.