document-generator

# Document Generator Process Core process skill for generating professional business documents. Defines the end-to-end workflow from user request to delivered file. ## Generation Workflow ### Step 1: CLASSIFY Determine the document type from the user's request: | Keywords | Document Type | Default Format | |----------|--------------|----------------| | proposal, bid, offer, pitch | **proposal** | DOCX | | invoice, bill, receipt | **invoice** | PDF | | estimate, quotation, quote, pricing | **estimate** | PDF | | report, analysis, findings, research | **report** | DOCX | | presentation, slides, deck, pitch deck | **presentation** | PPTX | | contract, agreement, MSA, terms | **contract** | DOCX | | NDA, non-disclosure, confidentiality | **nda** | PDF | | act, certificate of completion | **act** | PDF | | convert, transform, export | **conversion** | varies | ### Step 1.5: CHECK USER PREFERENCES (MANDATORY) **You MUST check preferences before any generation. Scripts will reject requests if preferences are missing.** ```bash cat ~/.document-generator/preferences.json 2>/dev/null ``` - **File missing** → **STOP. Run the onboarding interview** (see **user-preferences** skill) before proceeding. - **File exists** → load style preset, default company, language, and currency to pre-fill fields ### Step 2: GATHER Collect required data from the user. Each document type has specific required and optional fields. Refer to the **document-templates** skill for complete field checklists per document type. It defines exactly which fields are required, which are optional, and the expected format for each. If user preferences are loaded, pre-fill company info, currency, and language from the stored profile. Only ask for fields that aren't already known. ### Step 3: SELECT FORMAT & ENGINE | Type | Format | Engine | Script | Notes | |------|--------|--------|--------|-------| | Proposal | DOCX | docx-js | `generate_docx.js` | Default. Styled by the plugin (fonts, colours, cover page) | | Proposal | DOCX | pandoc | `generate_docx.js` (engine: "pandoc") | Same HTML as the PDF, but pandoc drops the CSS: structure only, look from `assets/reference.docx`. Requires pandoc | | Proposal (final) | PDF | playwright | `generate_pdf.js` | Full-page cover, TOC, sections | | Invoice | PDF | playwright | `generate_pdf.js` | Always PDF | | Report | DOCX | docx-js | `generate_docx.js` | Default. Styled by the plugin (fonts, colours, cover page) | | Report (final) | PDF | playwright | `generate_pdf.js` | Full-page cover, TOC, sections | | Presentation | PPTX | pptxgenjs | `generate_pptx.js` | Always PPTX | | Contract | DOCX | docx-js | `generate_docx.js` | Default for editable contracts | | Contract (final) | PDF | playwright | `generate_pdf.js` | Legal-grade formatting | | Estimate | PDF | playwright | `generate_pdf.js` | Phase-based pricing, timeline | | NDA | PDF | playwright | `generate_pdf.js` | Mutual or unilateral | | Act / Certificate | PDF | playwright | `generate_pdf.js` | Multi-language support | **Engine selection for DOCX:** - `docx-js` (default): Always works, full feature support (images, complex tables, headers/footers) - `pandoc`: Converts the same HTML templates to DOCX, but pandoc ignores CSS (inline `style` and `<style>` alike). Only the structure (headings, lists, tables) matches the PDF; fonts, colours and spacing come from `assets/reference.docx`. It does **not** look like the PDF — do not promise that. Set `"engine": "pandoc"` in input JSON. Requires pandoc installed (`which pandoc`). ### Step 4: BUILD JSON Input 1. Read the template file from `${CLAUDE_PLUGIN_ROOT}/templates/{type}_template.json` 2. Merge user data into the template structure 3. Set `outputPath` to: `{cwd}/{type}_{sanitized_title}_{YYYY-MM-DD}.{ext}` 4. Write the complete JSON input to a temp file: `{cwd}/.doc_input.json` **Input JSON structure:** ```jsonc { "type": "proposal|invoice|estimate|report|contract|nda|act", "engine": "playwright|pdfkit", "outputPath": "./proposal_acme_2026-03-17.docx", "template": { "...from template file..." }, "data": { "...user provided data..." } } ``` ### Step 5: GENERATE Run the appropriate script via Bash (`${CLAUDE_PLUGIN_ROOT}` is the plugin root; Claude Code substitutes the installed path — do not search for the plugin or `cd` into it): ```bash node "${CLAUDE_PLUGIN_ROOT}/scripts/generate_docx.js" /absolute/path/to/.doc_input.json ``` **Check dependencies first (use the centralized checker):** ```bash node "${CLAUDE_PLUGIN_ROOT}/scripts/check_deps.js" ``` If `ready: false`, show what's missing and ask user permission to install. **Script output** (JSON to stdout): - Success: `{ "success": true, "outputPath": "/abs/path/to/file.docx", "size": 12345 }` - Failure: `{ "success": false, "error": "description" }` ### Step 6: DELIVER 1. Parse the script's JSON output 2. Confirm the file exists and report its path and size 3. Offer follow-up actions: "Want me to convert this to PDF?" or "Want to make changes?" ## Format Conversion For converting between formats, use the pandoc wrapper: ```bash "${CLAUDE_PLUGIN_ROOT}/scripts/convert.sh" input.md output.pdf ``` Supported conversions: MD->PDF, MD->DOCX, MD->HTML, DOCX->PDF, DOCX->MD, HTML->PDF, HTML->DOCX, MD->PPTX ## Error Handling | Error | Resolution | |-------|-----------| | `warning: ONBOARDING_NOT_DONE` | Document generated with defaults. Offer to run `/document-generator-ops:setup` for custom styling | | Module not found (docx, playwright, etc.) | A marketplace install ships the npm modules (Claude Code installs them from `package-lock.json`). If the plugin is loaded in place (`--plugin-dir`), run `npm ci` in the plugin directory | | Playwright browser missing (`playwright_browsers` in `check_deps.js` output) | `npx playwright install chromium` — the browser goes to the global Playwright cache and survives plugin updates. Without a browser, `"engine": "pdfkit"` still makes a simple PDF | | pandoc not installed | `brew install pandoc` (macOS) or `apt install pandoc` (Linux) | | No PDF engine for pandoc | `brew install weasyprint` / `sudo apt install weasyprint` (recommended), or Typst (`brew install typst`); `pip install weasyprint` also works where pip is not locked down (PEP 668). `wkhtmltopdf` is deprecated in pandoc and is not used | | Playwright browser launch failed | Add `--no-sandbox` flag (already in script) | | Output file not created | Check script stderr, verify output directory exists | | Margin format mismatch | Scripts auto-normalize (twips ↔ CSS units) via `utils.js` | ## File Naming Convention Pattern: `{type}_{title}_{date}.{ext}` Sanitize title: lowercase, replace spaces with underscores, remove special characters, max 50 chars. Examples: - `proposal_acme_consulting_2026-03-17.docx` - `invoice_inv-001_2026-03-17.pdf` - `report_q1_analysis_2026-03-17.docx` - `presentation_product_launch_2026-03-17.pptx` - `contract_service_agreement_2026-03-17.docx` - `act_akt-001_2026-03-17.pdf`