restai
apocas/restai/CLAUDE.md
# CLAUDE.md This file provides guidance to Claude Code (claude.ai/code) when working with code
How real projects brief Claude Code: the commands, conventions and warnings they put in CLAUDE.md.
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
apocas/restai/CLAUDE.md
# CLAUDE.md This file provides guidance to Claude Code (claude.ai/code) when working with code
datadrivenconstruction/cad2data-Revit-IFC-DWG-DGN-pipeline-with-conversion-validation-qto/AI_AGENTS_INSTRUCTIONS/CLAUDE.md
Execute converters directly from terminal 3. **Process outputs**: Work with XLSX, DAE, IFC, HTML, PDF files ## Converters & CLI Syntax All converters are available for **Windows** (.exe) and **Linux
drasimwagan/mdv/CLAUDE.md
line charts accept `yFormat=currency|percent|thousands`. - **Exports**: self-contained HTML (`mdv render`) and PDF (`mdv export --pdf`, via puppeteer). ## Architecture Monorepo using npm workspaces: - `packages/mdv-core`: the parser (markdown
Piebald-AI/gemini-cli-desktop/CLAUDE.md
persistent width settings - **Cross-platform support**: Windows, macOS, and Linux compatibility - **File viewing support**: PDF, Excel, image, and text file viewers with syntax highlighting - **Advanced search**: Full-text search across
MIC-DKFZ/nnInteractive/CLAUDE.md
Code formatting black nnInteractive/ # Linting ruff check nnInteractive/ --fix # Spell checking codespell --skip='.git,*.pdf,*.svg' # Pre-commit hooks (after: pre-commit install) pre-commit run --all-files ``` There
trailofbits/skills-curated/CLAUDE.md
references/, quality gates, command interface | | **Advanced** | [culture-index](https://github.com/trailofbits/skills/tree/main/plugins/culture-index) | Scripts, workflows/, templates/, PDF extraction, multiple entry points | **When in doubt, copy one of these and adapt it.** **Deep
daloopa/investing/CLAUDE.md
TICKER` — Cash conversion cycle, earnings quality & working capital deep-dive ### Investment Deliverables (.docx, .xlsx, .pdf) - `/research-note TICKER` — Generate professional Word research note - `/build-model TICKER` — Build multi-tab Excel financial model
jhljx/RecSys-Industrial-Book/CLAUDE.md
claude/skills/recsys-engineering/SKILL.md` The PDFs are authoritative. Build only the ignored local index; never commit extracted PDF text or a prose-equivalent Markdown book
Siddhesh2377/ToolNeuron/CLAUDE.md
# ToolNeuron — Repo Guide Project memory for this repo. **When you change anything that affects
vorojar/Folio-OCR/CLAUDE.md
Chinese OCR prompt: 识别正文 + 跳过页眉页脚 + 表格输出为 Markdown/HTML - Auto-strips ```` ```markdown ``` ```` fences from model output - PDF → PNG via PyMuPDF at 2x resolution - Upload returns SSE stream (`init` → `page` × N → `done
duolahypercho/fusion-fable/CLAUDE.md
best practices for creating different document types (a docx skill for Word documents, a PDF skill for creating/filling PDFs, etc). These encode hard-won trial-and-error about producing professional
guiyumin/vget/CLAUDE.md
defines supported media types: - `MediaTypeVideo` - Video files (Twitter, etc.) - `MediaTypeAudio` - Audio files (podcasts) - `MediaTypePDF` - PDF documents - `MediaTypeEPUB` - EPUB ebooks - `MediaTypeMOBI` - MOBI ebooks - `MediaTypeAZW` - AZW ebooks - `MediaTypeUnknown` - Fallback (treated as video
Akylas/oss-weather/.claude/CLAUDE.md
lives under `app/`: - `app/components/` — all `.svelte` screens/components/modals, grouped by feature (`camera`, `edit`, `list`, `ocr`, `pdf`, `pkpass`, `qrcode`, `security`, `settings`, `view`, `common`, `widgets`). Platform-specific logic uses `.android.ts` / `.ios.ts`. - `app/services/` — domain
linuz90/claude-telegram-bot/CLAUDE.md
documents - **`photo.ts`** - Image analysis with media group buffering (1s timeout for albums) - **`document.ts`** - PDF extraction (pdftotext CLI), text files, archives, routes audio files to `audio.ts` - **`video.ts`** - Video messages and video
neosun100/DeepSeek-OCR-WebUI/frontend/CLAUDE.md
party OCR APIs**. The tool converts scanned images/multi-page PDFs into: - Markdown - DOCX - Text+Image PDF All processing except OCR API calls must run **entirely in the browser**. --- ## Core Constraints (VERY
benjamin-awd/monopoly/CLAUDE.md
passed, failed, or **skipped** | | `/add-bank` | Add a bank or statement type from a sample PDF | | `/debug-statement` | Diagnose a single PDF that fails to parse | | `/adr` | Record an architectural decision
shannhk/llm-wikid/CLAUDE.md
video and download images | | Web URL / Reddit URL | `scrapling` | Page content, title, author, date | | PDF file (.pdf) | Read directly / `summarize` CLI | Full text extraction | | Bookmark dump (list of URLs) | Parse
Kocoro-lab/ShanClaw/CLAUDE.md
# Kocoro — Project Guide ## What This Is Go CLI tool (`shan`) — the runtime for Shannon
ying-ge/FigureYa/CLAUDE.md
knitted report ├── easy_input_*.csv # Standardized input data files ├── example*.png # Style reference images ├── *.pdf # Vector graphic outputs └── output_*.txt # Results and tables ``` ### Key Architectural Patterns **1. Self-Contained Analysis
anthropics/healthcare/plugins/healthcare/skills/fraud-detection/scripts/CLAUDE.md
runs first (e.g. `fixedWidthToCsv()`), writes a clean headered CSV, then `ingest:` reads it normally. | | `pdf` | prose policy doc (NCCI manual, MLN article) | curl → `pdftotext -layout` → `policy/ .txt`. No DuckDB table
A file Claude Code reads at the start of every session. It holds the commands, conventions and warnings the agent needs for this project.
At the repository root. Claude Code also reads CLAUDE.md files in subdirectories when it works there.
Build and test commands, the project's layout, conventions that aren't obvious from the code, and mistakes to avoid. Short files tend to work better than long ones.
Claude Code reads CLAUDE.md; most other agents read AGENTS.md. Many projects keep one and point the other at it.