Implement the md-lines hook: one sentence per line in Markdown

[CodyCBakerPhD]

What this is

A general pre-commit hook grown out of scripts/check_sentence_per_line.py in nwb-benchmarks-paper. Consumers add it with repo: + rev:; it also installs as the md-lines command. Python 3.13 or newer; consumers whose pre-commit runs under an older interpreter add language_version: python3.13 to the hook entry.

- repo: https://github.com/CodyCBakerPhD/md-lines-pre-commit
  rev: v0.1.0   # tag to create once this merges
  hooks:
    - id: md-lines

How it differs from the original script

• Block structure comes from a CommonMark parser (markdown-it-py, plus tables and strikethrough). Only paragraph lines are reflowed, including paragraphs inside list items and block quotes. Headings, code, tables, HTML, link reference definitions, thematic breaks and YAML/TOML front matter are never touched. This fixes the two things the script got wrong on the runner README: a wrapped line starting with a Markdown link was never joined, and clauses after ;/: were left as fragments.
• Prefix handling: new lines inside block quotes get the > marker, inside list items the item's indentation (also for nested lists and loose items). Lazy block quote continuations are normalized.
• Sentence detection ignores punctuation in inline code, link destinations, autolinks, inline HTML, bare URLs and $math$; respects abbreviations (built-in list, --abbreviations, --abbreviations-file, --no-default-abbreviations), initials, dotted abbreviations, enumerators, bold lead-ins (**Note.**) and unbalanced bold markers, brackets or parentheses.
• Options: --check, --diff, --split-lists, --keep-breaks-after (default .!?; add :; to keep the original script's behavior of breaking after colons and semicolons), -q.
• Fidelity: CRLF, BOM and a missing final newline are preserved; the output is a fixed point (running again changes nothing). Exit codes 0 / 1 / 2.

Layout

• src/md_lines/schema/md_lines.yaml: the data model (Options, Problem, Result, Paragraph) as a LinkML schema.
• src/md_lines/models.py: generated from the schema with gen-python --no-metadata; a test fails when the two drift apart. Excluded from the formatters and from coverage.
• src/md_lines/globals.py: the abbreviation list, the parser, the regular expressions and DEFAULT_OPTIONS.
• src/md_lines/reflow.py: functions only. src/md_lines/cli.py: the command. src/md_lines/_version.py: reads the version from the installed metadata, so pyproject.toml is the only place it is written.
• AGENTS.md ported from historia and applied: relative internal imports, whole-module external imports, keyword-only arguments on multi-input functions, ai_generated test marker, CHANGELOG.md.

Repository checks

pre-commit after historia: pre-commit-hooks, check-github-workflows, black, ruff (lint plus isort rules, --fix), codespell with the en-GB_to_en-US dictionary so the text stays in American English, and this repository's own md-lines hook on its Markdown. pre-commit.ci runs the hooks on every pull request, except the md-lines hook, which is language: system and skipped there. The Continuous Integration workflow runs the tests on the newest Python with coverage uploaded to Codecov (using the CODECOV_TOKEN repository secret), runs that one hook, and installs the hook the way a consumer does with pre-commit try-repo.

Validation

• 52 tests (pytest), covering joins, splits, lists (kept, split, nested, loose, lazy), block quotes, every protected block type, front matter, hard breaks, inline masking, abbreviations, bold labels, CRLF/BOM, CLI modes and exit codes, the console entry point, and the schema/model agreement. 97% line coverage.
• Compared against the original script on the runner repo's README: identical output except for the fragments the script left behind, which md-lines joins.
• pre-commit run --all-files and pre-commit try-repo . md-lines --files README.md pass locally on Python 3.13.

After merging

Create the v0.1.0 tag so consumers can pin rev: v0.1.0. No LICENSE file is included; that choice is yours.

🤖 Generated with Claude Code

https://claude.ai/code/session_01ToiYPiSdU8NM9h1UmLgsq8

[CodyCBakerPhD]

The tests check is red on 2cc5030 only because of the new Codecov upload step: the tests themselves passed, then codecov-action logged "Token length: 0" and "Token required - not valid tokenless upload". The upload is set to fail CI on error, as in historia.

Nothing in the branch can fix this. The repository needs a CODECOV_TOKEN secret (Settings, Secrets and variables, Actions) holding this repository's upload token from Codecov. Re-running the job without it would fail the same way, so I have not re-run it. Once the secret exists, the next push or a manual re-run goes green.

---

Generated by Claude Code

[codecov-commenter]

Welcome to Codecov 🎉

Once you merge this PR into your default branch, you're all set! Codecov will compare coverage reports and display results in all future pull requests.

Thanks for integrating Codecov - We've got you covered ☂️