slopscore

https://jman4162.github.io/slopscore/
Admin
slopscore is a transparent linter for AI-slop writing patterns in essays, blog posts, Markdown,...
Tokens:5,798
Snippets:42
Trust Score:9.7
Update:1 month ago
Show doc for...
Tokens:
Raw
### Install slopscore-lint with pip extras

Source: https://jman4162.github.io/slopscore

Install the slopscore-lint package. The base install is lean and rule-based; the [report] extra adds HTML reports, [nlp] adds spaCy precision, and [all] installs everything.

```bash
pip install slopscore-lint            # lean, rule-based core
pip install "slopscore-lint[report]"  # + HTML reports
pip install "slopscore-lint[nlp]"     # + spaCy precision
pip install "slopscore-lint[all]"     # everything
```

--------------------------------

### slopscore.toml configuration example

Source: https://jman4162.github.io/slopscore/configuration

Example configuration for slopscore, placed in slopscore.toml or under [tool.slopscore] in pyproject.toml. Sets profile, strictness, scorer, and CI gate options; the exclude list replaces the built-in defaults. CLI flags override these file values.

```toml
# slopscore.toml  (or under [tool.slopscore] in pyproject.toml)
profile = "blog"            # blog | essay | academic | marketing | technical | social
strictness = "conservative" # conservative | balanced | sensitive
scorer = "rules"            # rules (default) | ml
min_reliable_words = 300
suggest = false             # include opt-in rewrite suggestions
broad = false               # include the opt-in broad rule tier (higher false-positive rate)

# CI gate. Either condition exits 1. Abstained documents (under 100 words, non-English) never
# fail on score, because their number is not a confident label.
fail_on = "none"            # none | low | medium | high: any rule hit at this severity fails
score_threshold = 50        # any non-abstained document at or above this score fails

# Batch walker filters (directory and multi-file scans). `exclude` REPLACES the defaults
# (node_modules, .venv, venv, .git, vendor, dist, build, __pycache__, .tox, .mypy_cache,
# .ruff_cache, site); patterns are root-relative globs or bare directory names.
include = ["docs/*", "*.md"]
exclude = ["node_modules", "build", "*.generated.md"]

# Turn a whole dimension off (it then scores 0 and emits no findings):
disabled_dimensions = ["formatting_tells"]

# Turn individual rules off, or override their severity:
disabled_rules = ["FORMULAIC_IN_CONCLUSION"]
rule_severity = { "COPULA_SERVES_AS" = "low" }

```

--------------------------------

### Scan prose inside code with slopscore-lint scan

Source: https://jman4162.github.io/slopscore/code-and-fairness

Use the scan command to extract prose from source files and check it for slop. The first example scans a single file, the second recursively scans a package, and the third scans a Markdown file with a fail-on threshold of high. Offsets in the report index the extracted prose, so SARIF and HTML line numbers are relative to that prose, not the raw source bytes.

```shell
slopscore-lint scan src/app.py            # docstrings + comments in one file
slopscore-lint scan src/ --recursive      # across a package
slopscore-lint scan README.md --fail-on high

```

--------------------------------

### Audit fairness with slopscore-lint fairness

Source: https://jman4162.github.io/slopscore/code-and-fairness

Use the fairness command to audit rule false-positive rates on plain and non-native English. The first example reports per-rule false-positive rates on the simple_english and non_native slices; the second adds a threshold of 0.1 to flag rules that fire over 10 percent. Document-level false positives are 0 percent on both slices, and two performative_candor rules fire at the rule level but do not push a document over the threshold.

```shell
slopscore-lint fairness                  # per-rule false-positive rate on plain/ESL text
slopscore-lint fairness --threshold 0.1  # flag rules over 10 percent

```

--------------------------------

### Scan files with slopscore-lint CLI

Source: https://jman4162.github.io/slopscore

Run slopscore-lint scan on a file or directory. Use --recursive --fail-on high for CI gating, --format sarif -o out.sarif for GitHub code scanning, --suggest for rewrite suggestions, and --broad for the broad tier (jargon + bare weasel words).

```bash
slopscore-lint scan post.md
slopscore-lint scan ./content --recursive --fail-on high      # CI gate
slopscore-lint scan post.md --format sarif -o out.sarif       # GitHub code scanning
slopscore-lint scan post.md --suggest                         # opt-in rewrite suggestions
slopscore-lint scan post.md --broad                           # opt-in broad tier (jargon + bare weasel words)
```

--------------------------------

### Record and Scan with Baseline

Source: https://jman4162.github.io/slopscore/baseline

Record current findings to a baseline file and use that file in CI to gate only on new findings.

```bash
# 1. Record current findings (commit the result).
slopscore-lint baseline ./content --recursive -o .slopscore-baseline.json

# 2. In CI, gate only on findings not in the baseline.
slopscore-lint scan ./content --recursive \
  --baseline-file .slopscore-baseline.json --fail-on-new
```

--------------------------------

### Inline suppression with slopscore-disable comments

Source: https://jman4162.github.io/slopscore/suppression

Use these HTML comments to suppress SlopScore findings inline. `disable-next-line` silences the next line, `disable-line` silences the current line, `disable`/`enable` wrap a block, and `disable-file` suppresses the entire file. Names are comma-separated rule IDs or dimension names; omitting them suppresses everything. Unknown names produce a warning.

```html
<!-- slopscore-disable-next-line unsupported_claims -->
Everyone knows this is the future.

This line is fine. <!-- slopscore-disable-line COPULA_SERVES_AS -->

<!-- slopscore-disable significance_inflation -->
A block where significance-inflation findings are ignored …
<!-- slopscore-enable significance_inflation -->

<!-- slopscore-disable-file -->
Suppress every finding in this file (e.g. generated content).

```

--------------------------------

### Triage a long document with slopscore-lint scan --by-paragraph

Source: https://jman4162.github.io/slopscore/code-and-fairness

Use the --by-paragraph flag with scan to score each paragraph and print them worst-first, so you can find the section to fix in a long document. Short paragraphs abstain, so treat this as triage, not a per-paragraph verdict.

```shell
slopscore-lint scan article.md --by-paragraph

```
Tokens:
Raw