Skip to content

pre-commit.md

Latest commit

 

History

History
97 lines (70 loc) · 2.9 KB

pre-commit.md

File metadata and controls

97 lines (70 loc) · 2.9 KB
Raw

Gate commits on the AI-tell score

The repo ships two pre-commit hooks. Both run the local metrics CLI, which has no dependencies, makes no network calls, and reads no credentials. Your text never leaves the machine.

The two hooks

id Scans Use it when
humanizer-scan-staged Only the .md, .markdown and .txt files in the commit Almost always. Fast, and it fails on what you actually changed.
humanizer-scan The whole repository, every commit You want a standing floor for the entire tree, not just the diff.

Both fail the commit when a file scores above 40 on the 0-100 AI-tell scale.

Setup

Add this to .pre-commit-config.yaml in your own repo:

repos:
  - repo: https://github.com/Aboudjem/humanizer-skill
    rev: v0.7.0
    hooks:
      - id: humanizer-scan-staged

Then:

pre-commit install
pre-commit run --all-files

rev is a git tag on this repository, not the CLI's own package version. Pick a released tag so your gate does not move under you.

Choosing a threshold

The hooks ship with --fail-above 40, which is the top of the "Mostly human" band. Run the CLI on a file you are happy with and use the number it prints:

node cli/index.js score examples/blog-post/after.md
examples/blog-post/after.md
Score: 21/100  (Mostly human)
  Signal breakdown (points out of 100):
    lexical     19.1   weight 0.4   vocabTellDensity 0.0167
    repetition  1.9    weight 0.14  trigramRepetition 0.024
    burstiness  0.0    weight 0.28  burstiness 1.291
    diversity   0.0    weight 0.18  mattr 0.887

The breakdown says where the points came from, so you can see whether a file is failing on vocabulary, on flat sentence lengths, or on repetition.

To override the threshold, pass your own args in .pre-commit-config.yaml:

      - id: humanizer-scan-staged
        args: [scan, --fail-above, '30']

The bands are: 0-20 Pristine, 21-40 Mostly human, 41-60 Mixed, 61-80 AI-leaning, 81-100 Pure AI smell. Prose written by a person on a normal day lands in the first two bands. A threshold below 20 will fail honest writing.

Exit codes

Code Meaning
0 Every scanned file is at or under the threshold
1 At least one file is above it
2 Usage error, or a path that does not exist

A path that cannot be read is an error, not a skip, so a typo in a config fails loudly instead of quietly scanning nothing.

Running it in CI instead

The same gate works without pre-commit:

node cli/index.js scan docs --fail-above 40

Or hold a committed baseline and fail only on files that got worse:

node cli/index.js scan docs --baseline .humanizer-baseline.json --write-baseline
node cli/index.js scan docs --baseline .humanizer-baseline.json --fail-on-regression

There is also a ready-made composite action at .github/actions/humanizer-gate/action.yml.