mdselect 0.2.0

Select Markdown sections by heading path

mdselect

中文说明

Select Markdown sections by heading path.

mdselect is a small, zero-runtime-dependency command-line tool for inspecting a Markdown outline and extracting complete sections without losing their original formatting.

Why mdselect?

Line-oriented tools such as grep can find a heading, but they do not know where that section ends. A full Markdown AST library understands much more of the format, but usually requires writing a program and choosing rendering semantics.

mdselect fills the gap:

• query the heading hierarchy directly from the shell;
• return a matched heading together with its body and descendants;
• preserve the original Markdown content without parsing and re-rendering it;
• use regular expressions in a compact heading path expression;
• run with only the Python standard library.

Installation

mdselect requires Python 3.10 or newer.

pipx install mdselect
# or
uv tool install mdselect
# or
python -m pip install mdselect

To install a development checkout:

python -m pip install -e ".[test]"

Codex plugin

This repository also provides the recall-markdown-memory skill as an installable Codex plugin. The skill uses the mdselect executable from the task environment; installing the plugin does not install Python or the PyPI package automatically.

Install the CLI first and verify that it is on PATH:

pipx install mdselect
mdselect --version

Then add the repository marketplace and install the plugin:

codex plugin marketplace add stevenybw/mdselect
codex plugin add mdselect@mdselect

Start a new Codex session after installation. Invoke the skill explicitly with $mdselect:recall-markdown-memory, or ask Codex to recall a decision, preference, or other context from Markdown memory and let the skill trigger automatically.

The first release targets shell-enabled Codex CLI, IDE, and local desktop tasks. For a Codex Cloud environment, add CLI installation to its setup script, for example:

python -m pip install mdselect

The skill is read-only. It searches only user-specified or project-configured memory locations, plus the conventional MEMORY.md, memory/, and .agents/memory/ paths; it does not scan every Markdown file by default.

Quick start

Given handbook.md:

# Engineering Handbook

## Getting Started

Introductory material.

### Local Setup

Create a virtual environment.

## Operations

Operational notes.

Print its outline:

mdselect outline handbook.md

# Engineering Handbook
  ## Getting Started
    ### Local Setup
  ## Operations

Extract the complete Getting Started subtree:

mdselect select --mpath '/Engineering/Getting Started' handbook.md

## Getting Started

Introductory material.

### Local Setup

Create a virtual environment.

Add a trailing slash to omit the matched heading line:

mdselect select --mpath '/Engineering/Getting Started/' handbook.md

mpath syntax

An mpath is a slash-separated heading path:

Form	Meaning
/Guide/Install	Absolute query: the first segment only matches level-one headings.
Install	Relative query: the first segment matches headings at any level.
/^Guide$/^Install$/	Each segment is a Python regular expression; anchors make matches exact. The trailing slash omits the matched heading.
/	Return the original document unchanged.

The complete rules are:

1. Each non-empty segment is compiled as a Python regular expression and matched with re.search.
2. A leading slash makes the query absolute: its first segment is tested only against level-one (#) headings.
3. Without a leading slash, the first segment is tested against every heading.
4. Every later segment is tested against descendants of the current matches, at any depth.
5. A trailing slash excludes the final matched heading line while keeping its body and descendants.
6. /, //, and an empty expression are no-ops that return the original document.
7. A slash cannot appear inside a segment.
8. Multiple matches retain document order and are separated by one blank line.
9. When both an ancestor and its descendant match, only the ancestor is rendered as a result because its subtree already contains the descendant.

mdselect select -h prints these rules without requiring the README.

Exit status

Situation	Status
Successful outline or selection	0
No matching section	0, with empty standard output
Invalid arguments or regular expression	2
File cannot be read	non-zero

The no-match behavior is intentional so selection can be used in shell pipelines without treating an optional section as a command failure.

Supported Markdown and boundaries

• ATX headings from # through ###### are supported when the marker starts in column one and is followed by whitespace.
• Optional closing hash markers are removed from the heading title used for matching.
• Backtick and tilde fenced code blocks are recognized; heading-like lines inside a fence are ignored.
• Selected content is rendered from the original lines, preserving tables, blockquotes, blank lines, inline markup, and fences.
• Files are read as UTF-8.
• Setext headings are not supported.
• mdselect does not interpret front matter, links, inline Markdown, or HTML as an AST.

Comparison

Tool	Best for	Structural section extraction
grep / rg	Finding matching lines across files	No
Markdown AST library	Full parsing, transformation, and custom rendering	Yes, with application code
mdselect	Shell-friendly outline and heading-subtree selection	Yes, directly from the CLI

Use grep or rg to discover candidate files, then use mdselect when the unit you need is a Markdown section rather than a line.

Python API

from mdselect import outline, select

print(outline(markdown_text))
section = select(markdown_text, "/Guide/Install")

The package also exports Section, parse, render_subtree, and InvalidPattern.

Contributing and security

See CONTRIBUTING.md for development and pull request guidance. Report security issues privately as described in SECURITY.md.

License

MIT