Agent Skills

all2md ships with a pre-built agent skill — a structured instruction file that teaches AI coding assistants (Claude Code, Cursor, Windsurf, etc.) how to use all2md effectively. It follows Anthropic’s progressive-disclosure pattern: a lean SKILL.md overview routes to focused, on-demand reference guides for each document workflow.

Overview 

An agent skill is a directory containing a SKILL.md file with YAML frontmatter (name + description). Agents load the description at startup, read the full SKILL.md when a task matches, and pull in the bundled reference files only as needed. This keeps the always-loaded footprint small while making deep, per-task detail available on demand.

all2md bundles a single all2md skill whose per-task guides live in a references/ directory:

all2md/
├── SKILL.md                # overview + index (routes to the references below)
└── references/
    ├── read.md             # read / extract text & tables from any document
    ├── convert.md          # convert between formats (PDF→DOCX, HTML→PDF, …)
    ├── generate.md         # create documents from Markdown (DOCX, PDF, PPTX, EPUB, sites)
    ├── grep.md             # pattern matching inside documents (grep for PDFs)
    ├── search.md           # ranked / semantic search across collections
    └── diff.md             # compare any two documents regardless of format

Installing Skills 

Use the install-skills CLI command to copy the bundled skill (and its references) to your agent’s skills directory:

# Install to default location
# (uses ./.agents/skills/ if it exists, otherwise ~/.agents/skills/)
all2md install-skills

# Install to local project directory
all2md install-skills --local

# Install to global home directory
all2md install-skills --global

# Install to a custom directory
all2md install-skills --target /path/to/skills

# Overwrite existing skills
all2md install-skills --force

# List bundled skills without installing
all2md install-skills --list

# Remove previously installed skills
all2md install-skills --uninstall

After installation, the all2md/ skill directory contains SKILL.md plus the references/ guides that agents discover and load on demand.

Reading the guide without installing 

The same content is available straight from the CLI — handy when an agent is driving all2md from a terminal and hasn’t installed the skill:

# Print the overview plus every reference, concatenated
all2md llm-help

# Print a single topic
all2md llm-help read        # or convert, generate, grep, search, diff, overview

# List available topics
all2md llm-help --list

Reference Topics 

read 

The most common agent use case — “read this document and give me the text.”

Triggers: Reading PDFs, Word docs, PowerPoint, Excel, HTML, emails, images, or any other document format; extracting text or tables; parsing document structure.

Key capabilities covered:

Basic conversion: all2md document.pdf, all2md report.docx -o report.md
Stdin: cat doc.pdf | all2md -
Format-specific options: --pdf-pages, --pdf-table-detection-mode, --pdf-ocr-enabled, --docx-include-comments, --html-extract-title, --eml-attachment-mode
Section extraction: --extract "Chapter 3", --outline; line-range navigation: --outline --line-numbers then --extract line:42-87
Batch: -r, --parallel, --collate
Python API: to_markdown(), to_ast(), parser options

convert 

Any-to-any format conversion beyond just reading to Markdown.

Triggers: Converting PDF to Word, HTML to DOCX, DOCX to PDF, any-to-any format conversion, changing file formats.

Key capabilities covered:

Common conversions: all2md input.pdf --output-format docx -o output.docx
AST transforms during conversion: --transform remove-images
Format listing: all2md list-formats, all2md list-transforms
Python API: convert(), from_markdown(), to_ast() / from_ast(), apply()

generate 

Create documents from scratch using Markdown as input.

Triggers: Creating Word documents, generating PDFs, making slides, producing EPUB ebooks, building static sites, ArXiv packaging.

Key capabilities covered:

Document generation: all2md report.md --output-format docx -o report.docx
Per-format options: DOCX templates (--docx-renderer-template-path), PDF page size (--pdf-renderer-page-size), HTML standalone (default; --html-renderer-no-standalone for a fragment), EPUB
Template rendering: --output-format jinja --jinja-renderer-template-file template.html
Static sites: all2md generate-site ./docs --output-dir site --generator hugo
ArXiv: all2md arxiv paper.md -o submission.tar.gz
Python API: from_markdown(), renderer options, ArxivPackager

grep 

The agent equivalent of grep but for any document format.

Triggers: Searching inside PDFs, grepping Word documents, finding text in PowerPoint slides, pattern matching in non-plaintext files.

Key capabilities covered:

Basic grep: all2md grep "pattern" document.pdf
Multiple files: all2md grep "pattern" *.pdf ./docs/*.docx
Flags: -i, -n, -e/--regex, -B/-A/-C, -M
Rich output: --rich
Stdin: cat doc.pdf | all2md grep "pattern" -
Recursive: all2md grep "pattern" ./documents -r

search 

Advanced ranked search across document collections.

Triggers: Searching across many documents, semantic search, keyword ranking, building search indexes.

Key capabilities covered:

Search modes: --keyword (BM25), --vector (semantic), --hybrid (or --mode keyword|vector|hybrid|grep)
Result control: --top-k 5; grep-mode context with --mode grep -C 2
Persistent indexes: --index-dir ./search-index --persist
Output formats: default plain, --rich, --json
Chunk size: --chunk-size 500

diff 

Compare any two documents, regardless of format.

Triggers: Comparing documents, diffing files, version comparison, finding differences.

Key capabilities covered:

Basic diff: all2md diff original.docx modified.docx
Cross-format: all2md diff document.pdf document.docx
Output formats: default unified, --format html, --format json
Options: --ignore-whitespace, --context 5, -C 0
Granularity: --granularity block, sentence, word
Color: --color always, --color never

How Skills Work 

The bundled skill follows the Agent Skills specification:

Frontmatter — a YAML block with name and description (plus optional metadata)
Description — states what the skill does and when to use it, so the agent can decide when to load it
Body — a lean overview that links to the per-task references under references/
References — focused guides loaded on demand (progressive disclosure)

When an agent encounters a document-related task, it matches the user’s intent against the skill description, loads SKILL.md, and then reads only the reference file relevant to the task.

Customizing Skills 

After installing, you can edit SKILL.md or any file under references/ to add project-specific patterns or trim sections you don’t need. For example, if your project only uses PDF and DOCX, you could shorten references/read.md to reduce noise.

To reset to the defaults, run all2md install-skills --force.