Run Once
uvx --from slop-guard sg README.mdCatch slop before it lands in your docs.
slop-guard is a rule-based prose linter for formulaic AI
writing. It scores text from 0 to 100, points to the exact spans
that dragged the score down, and returns direct advice for the
rewrite.
No LLM judge. No API calls. No cost.
Install The CLI
uv tool install slop-guardAdd To Codex
codex mcp add slop-guard -- uvx slop-guardAdd To Claude Code
claude mcp add slop-guard -- uvx slop-guardFrom Slop To Clean
0 / 100
to
100 / 100
Features
Score
0-100 banding
Clean, light, moderate, heavy, or saturated. The score is compact enough for a terminal and sharp enough for a CI floor.
Matches
Exact spans
Violations come back with the matched text, surrounding context, and penalty. You can point at the problem instead of arguing about vibes.
Advice
Rewrite prompts
The output tells you what to cut and what kind of wording to replace it with, so the rewrite is faster than a blank restart.
Agent Workflow
Inside Codex or Claude
Add slop-guard as an MCP server and let the agent
call check_slop or
check_slop_file while it drafts docs, release
notes, or project updates.
CLI + CI
Run it locally or fail the build
Use sg on files, inline text, or stdin. Add a
threshold in CI and low-scoring prose exits non-zero before it
ships.
Custom Rules
Fit it to your own style
Train a JSONL rule config from positive and negative corpora, then point the CLI and MCP server at the fitted rules for your own house style.
Start Here
Keep the draft. Lose the filler.
Wire slop-guard into Codex or Claude Code, then let
the agent lint the writing before it ships.
codex mcp add slop-guard -- uvx slop-guard
claude mcp add slop-guard -- uvx slop-guard