My Skills Collection

A personal collection of Claude Code skills, versioned and shareable. Each skill lives in its own folder under skills/ and can be installed by copying it into your local Claude skills directory.

Repository structure

My-Skills-Collection/
├── LICENSE
├── README.md
└── skills/
    └── <skill-name>/
        ├── SKILL.md          # skill manifest (name + description frontmatter)
        └── scripts/          # optional helper scripts

Skills

Skill	Description
word-md-roundtrip	Round-trip between Microsoft Word (.docx) and a chapter-split, AI-context-friendly Markdown structure. PowerShell + Word COM + Pandoc; hierarchical folder split by token budget, navigation hub with WikiLinks, image extraction, and XML-bound placeholders (content controls). Not lossless — Markdown-representable content is preserved; see Limitations.

Installation

Claude Code discovers skills placed in your user skills directory. To install a skill from this collection, copy its folder there:

Windows (PowerShell):

Copy-Item -Recurse -Force `
  ".\skills\word-md-roundtrip" `
  "$env:USERPROFILE\.claude\skills\"

macOS / Linux:

cp -r ./skills/word-md-roundtrip ~/.claude/skills/

Once installed, the skill is invoked via /word-md-roundtrip or picked up automatically by Claude when a matching request is made.

Note: scripts are stored as UTF-8 with BOM so they parse correctly under both Windows PowerShell 5.1 and PowerShell 7. Preserve the byte content when copying (a plain file copy does this; avoid re-encoding round-trips).

Skill: word-md-roundtrip

Splits a Word document into a recursive Markdown folder structure (chapters are never cut mid-meaning, sized to a token budget), and merges it back into Word.

The round trip is not lossless: it faithfully preserves text, heading structure, tables, lists, images, foot-/endnotes and Markdown-representable formatting (plus separately tracked XML-bound placeholders), but everything that Markdown/Pandoc cannot represent is dropped or approximated — see Limitations.

Requirements: Windows · Microsoft Word (COM automation) · PowerShell 5.1 or 7 · Pandoc (auto-installed via winget if missing).

Quick usage (run from the installed skill's scripts/ folder):

# Word -> Markdown (non-destructive; creates "01 - <Title>/" next to the source)
pwsh -NoProfile -File .\Convert-WordToMarkdown.ps1 -Path "C:\Docs\Manual.docx"

# Markdown -> Word (overwrites the original; auto-backup in .backup\)
pwsh -NoProfile -File .\Convert-MarkdownToWord.ps1 -DocDir ".\01 - Manual"

# ...or write the result elsewhere, leaving the original untouched
pwsh -NoProfile -File .\Convert-MarkdownToWord.ps1 -DocDir ".\01 - Manual" -OutFile "C:\out\Manual.docx"

# End-to-end self-test (generates a sample document)
pwsh -NoProfile -File .\Test-Roundtrip.ps1

Limitations — not lossless

By deliberate scope, the round trip does not preserve:

• Word-only constructs: comments, tracked changes, headers/footers, Word fields, and live tables of contents. An auto-generated TOC chapter is intentionally discarded (it is regenerable in Word and only adds noise).
• Complex/custom styling: formatting is reduced to what Markdown + Pandoc can express. The original document is used as a Pandoc reference doc so the result resembles it, but exact layout, spacing, themes and advanced styles are approximated, not byte-preserved.
• Equations / formulas: best-effort via Pandoc; may degrade or be lost.
• Placeholders outside the body flow: content controls on cover pages, in text boxes, or in headers/footers are recovered into the hub fields: registry as metadata, but produce no {{token}} in the body and are therefore not re-inserted on the reverse pass (Pandoc does not export those regions).
• Title & names: the document title is taken from Word metadata — any typo in the source carries through. File/folder/hub names are stripped of Markdown formatting and newlines (the raw heading line is kept verbatim in the body, so formatting survives the round trip there).
• Token sizing is heuristic: chapter split sizes use an approximate token estimate, not an exact tokenizer.

Operational notes:

• Markdown → Word overwrites the original by default (a timestamped copy is written to .backup\). Use -NoOverwrite or -OutFile <path> to avoid this.
• [[WikiLinks]] / ![[embeds]] are non-standard Markdown understood only by this skill's reverse pass. Do not manually rename the generated folders or numbering, or the reverse merge will break.
• {{ }} is reserved for placeholders; avoid literal double curly braces in body text.

See skills/word-md-roundtrip/SKILL.md for the full workflow, parameters, placeholder/field model, and the complete list of documented scope limits.

Adding a new skill

Create skills/<skill-name>/ containing a SKILL.md with YAML frontmatter (name and description) and any supporting scripts/. Add a row to the Skills table above, then commit and push.

License

MIT © 2026 SZzip