Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Build a real, formatted Word (.docx) document — headings, styles, tables, TOC-ready. Use when asked to produce a Word doc, a .docx, a formatted report/contract/proposal/letter as an actual file (not markdown). Produces an actual .docx via a generated python-docx script with proper heading styles, body text, tables, and page structure. Requires a code-execution environment (Claude Code, the API code tool, or Claude.ai).
.claude/skills/mohitagw15856-word-document/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-08 | ✗→✓ | ▲ Improved | 12% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 240% | 0% |
| case-03 | ✓→✗ | ▼ Worse | -83% | 0% |
| case-09 | ✓→✓ | = Same ✓ | 63% | 0% |
| case-05 | ✓→✓ | = Same ✓ | 70% | 0% |
When someone needs an actual .docx — a report, proposal, contract, or formal letter they'll edit in Word — markdown won't do. This skill produces a real Word file by writing and running a python-docx script: proper heading styles (so the navigation pane and a TOC work), clean body text, tables, and page structure — a document that looks authored, not exported.
> Environment: produces a binary file, so it needs code execution — Claude Code, the > API code-execution tool, or Claude.ai. In the browser playground, the existing Word/PDF export > turns any skill's markdown into a document; this skill is for a built-to-spec .docx.
Ask for these only if they aren't already provided:
python-docx script that:Heading 1/2/3) — not bold body text — so the nav pane, cross-refs, and a generated TOC work..docx..docx file.Document-production practice (style-based formatting, structured headings, TOC-ready) implemented with python-docx.
This skill ships scripts/docx_tool.py — zero-dependency (stdlib zip+XML) production of real .docx files:
bash# Markdown-lite → Word (#/##/### headings, - bullets, 1. numbered, **bold**, *italic*) python3 scripts/docx_tool.py create out.docx --text-file doc.md # Fill {{placeholders}} through an existing .docx (body, headers, footers) — # handles Word splitting a placeholder across formatting runs python3 scripts/docx_tool.py fill template.docx out.docx --values '{"client":"Acme","date":"2026-07-03"}' # Verify what a .docx actually says (plain-text extraction) python3 scripts/docx_tool.py extract out.docx
Write the document first (per this skill), then create it as a real file. Honest limits: the markdown subset above with default styling; complex templates keep their formatting except in paragraphs where a placeholder spanned runs.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-09 | pass→pass | 15,413 | 16,963 | +10% | 1 | 1 | 0% | 1,965 | 3,209 | +63% | 0 | 0 | — |
case-01 | fail→fail | 56,936 | 15,324 | -73% | 1 | 1 | 0% | 8,268 | 1,274 | -85% | 0 | 0 | — |
case-02 | fail→fail | 47,260 | 15,291 | -68% | 1 | 1 | 0% | 8,260 | 1,233 | -85% | 0 | 0 | — |
case-03 | pass→fail | 40,420 | 16,921 | -58% | 1 | 1 | 0% | 8,257 | 1,403 | -83% | 0 | 0 | — |
case-04 | fail→fail | 24,826 | 53,754 | +117% | 1 | 1 | 0% | 4,801 | 9,220 | +92% | 0 | 0 | — |
case-05 | pass→pass | 14,256 | 15,076 | +6% | 1 | 1 | 0% | 1,605 | 2,734 | +70% | 0 | 0 | — |
case-06 | fail→fail | 22,417 | 28,207 | +26% | 1 | 1 | 0% | 4,714 | 1,950 | -59% | 0 | 0 | — |
case-07 | fail→fail | 49,286 | 67,363 | +37% | 1 | 1 | 0% | 8,266 | 9,238 | +12% | 0 | 0 | — |
case-08 | fail→pass | 101,670 | 54,132 | -47% | 1 | 1 | 0% | 8,253 | 9,225 | +12% | 0 | 0 | — |
case-10 | fail→fail | 55,238 | 62,210 | +13% | 1 | 1 | 0% | 8,227 | 9,199 | +12% | 0 | 0 | — |
case-11 | fail→pass | 22,591 | 41,009 | +82% | 1 | 1 | 0% | 2,643 | 8,997 | +240% | 0 | 0 | — |
case-12 | fail→fail | 22,091 | 58,000 | +163% | 1 | 1 | 0% | 3,787 | 9,206 | +143% | 0 | 0 | — |
case-13 | fail→fail | 37,130 | 23,031 | -38% | 1 | 1 | 0% | 7,280 | 6,133 | -16% | 0 | 0 | — |
case-14 | fail→fail | 14,157 | 21,448 | +52% | 1 | 1 | 0% | 1,599 | 1,679 | +5% | 0 | 0 | — |
case-15 | fail→fail | 23,314 | 22,535 | -3% | 1 | 1 | 0% | 3,665 | 1,721 | -53% | 0 | 0 | — |
case-16 | fail→fail | 45,595 | 5,350 | -88% | 1 | 1 | 0% | 8,227 | 1,243 | -85% | 0 | 0 | — |
case-17 | fail→fail | 53,705 | 38,577 | -28% | 1 | 1 | 0% | 5,454 | 7,304 | +34% | 0 | 0 | — |
case-18 | pass→pass | 21,201 | 42,958 | +103% | 1 | 1 | 0% | 2,313 | 8,707 | +276% | 0 | 0 | — |
case-19 | fail→fail | 57,244 | 17,940 | -69% | 1 | 1 | 0% | 8,221 | 1,623 | -80% | 0 | 0 | — |
case-20 | fail→fail | 45,620 | 18,314 | -60% | 1 | 1 | 0% | 8,232 | 3,270 | -60% | 0 | 0 | — |
case-21 | pass→pass | 10,599 | 4,007 | -62% | 1 | 1 | 0% | 962 | 1,640 | +70% | 0 | 0 | — |
case-22 | pass→pass | 43,027 | 46,161 | +7% | 1 | 1 | 0% | 8,240 | 7,450 | -10% | 0 | 0 | — |
case-23 | pass→pass | 12,807 | 4,720 | -63% | 1 | 1 | 0% | 1,313 | 2,004 | +53% | 0 | 0 | — |
case-24 | pass→pass | 42,862 | 39,087 | -9% | 1 | 1 | 0% | 8,219 | 8,623 | +5% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 24 cases were attempted, and 16 counted toward the lift figure. The other 8 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +4 percentage points is the difference between those two pass rates over the 16 comparable cases. 5 cases got worse with the skill loaded, and they are included in that figure.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.