プロジェクトディレクトリ下に置く。そのプロジェクトでのみ考慮される コーディングルールや、後述のメモリーバンクの設定などに利用します # General Constraints (for Programming in Python) ## Coding Style - Assume Python 3.10+ and use its features - Use the src layout: `src/<package>` for source code, `scripts` for scripts, `tests` for tests - Follow the latest guidelines (PEP 585) for type hints: e.g., use `list` `dict` rather than `typing.List`. - Use `logging.GetLogger` to instantiate logger and use it to print outputs ## Package Management - Use uv (https://github.com/astral-sh/uv) for package management - Initialize the repository with `uv init --lib` - Add/update dependencies with `uv add <package>` or `uv add --dev <package>` - Execute `uv run <python file>` when you run a python script. - Every time when you finish your task, execute `uvx ruff check --fix --extend-select I` to apply formatting ## Configuration Management - Use hydra (https://hydra.cc/) for configuration management - Define configuration files in `scripts/config/` in the same directory as the script file you are working. - Use `hydra.utils.instantiate` to instantiate class instances. - Set `hydra.job.chdir=False` so as not to change the working directory 設定の一例 一昨日仕様が変わりました
• Workspace Rules: そのワークスペースのみで有効 コーディングルールや、後述のメモリーバンクの設定などに利用します # General Constraints (for Programming in Python) ## Coding Style - Assume Python 3.10+ and use its features - Use the src layout: `src/<package>` for source code, `scripts` for scripts, `tests` for tests - Follow the latest guidelines (PEP 585) for type hints: e.g., use `list` `dict` rather than `typing.List`. - Use `logging.GetLogger` to instantiate logger and use it to print outputs ## Package Management - Use uv (https://github.com/astral-sh/uv) for package management - Initialize the repository with `uv init --lib` - Add/update dependencies with `uv add <package>` or `uv add --dev <package>` - Execute `uv run <python file>` when you run a python script. - Every time when you finish your task, execute `uvx ruff check --fix --extend-select I` to apply formatting ## Configuration Management - Use hydra (https://hydra.cc/) for configuration management - Define configuration files in `scripts/config/` in the same directory as the script file you are working. - Use `hydra.utils.instantiate` to instantiate class instances. - Set `hydra.job.chdir=False` so as not to change the working directory 設定の一例 ここから設定できます
bank を実行してメモリーバンクを作成します Planモードで以下のタスクを渡します（内容は適宜変更してください） Create README.md to explain the project. Specifically your tasks are: 1. Add type annotations and docstrings to all the classes and functions. 2. To the README.md, add the project overview, instructions for installation, and references for key classes and scripts. タスクの一例
bank を実行してメモリーバンクを作成します Planモードで以下のタスクを渡します（内容は適宜変更してください） Create a web demo that runs xxx (your model) on the browser ## Requirements - frontend: javascript, allowing users to upload images and displays outputs visually - backend: python flask, receiving uploaded images and run XXX Make sure both frontend and backend to be minimal and easy to read. タスクの一例
pushでoverleafと同期できてさらに良し Planモードで以下のタスクを渡します Review the manuscript (xxx.tex) and perform the following tasks section by section: - Fix all grammatical errors - Improve readability and clarity - Improve the consistency For each section or subsection, you have to first get the approval from the user about your revision plan. Make sure this is a scientific technical paper that will be submitted to robotics conferences. Avoid using too fancy words (eg innovative) and overstating phrases, and try to make the tone solid and objective. タスクの一例
-o https://raw.githubusercontent.com/nvm-sh/nvm/v0.40.3/install.sh | bash # Reload your shell or run: source ~/.bashrc # Install and use the latest LTS Node.js nvm install --lts nvm use --lts # Install Claude Code npm install -g @anthropic-ai/claude-code
all changes and: 1. Group related changes 2. Create multiple commits if needed 3. Use conventional commit format 4. Add emoji based on change type: - 🐛 for bug fixes - ✨ for new features - ♻ for refactoring - 📝 for documentation - 🎨 for style changes 5. Push to remote 6. Create PR if on feature branch
all changes and: 1. Group related changes 2. Create multiple commits if needed 3. Use conventional commit format 4. Add emoji based on change type: - 🐛 for bug fixes - ✨ for new features - ♻ for refactoring - 📝 for documentation - 🎨 for style changes 5. Push to remote 6. Create PR if on feature branch 適当にgitリポジトリ内でファイルを変更、同ディレクトリで Claude Codeを起動 /smart_pushで呼び出し
claude `--model` sonnet Clone https://github.com/ostris/ai-toolkit . Think and investigate the codebase and create markdown files to explain the structure. Include mermaid diagrams to explain how a main module runs in terms of data flow and module interaction. Additional diagrams to help understand the codebase is always welcome. Place the markdown files inside the repo’s structure/ directory. Sequence Diagram
of Thoughtに利用するトークン数を指定可能 複雑なタスク(推論を含む)についてはthinkをつけさせることで精度を向上できる think/think hard/ultrathinkキーワードを検出し、次の会話ターンでのみ思考トークンが設定される ✻ Thinking… The user is asking me to greet them like Gandalf. This is a creative request for a specific character greeting. I need to channel Gandalf's manner of speaking - formal, wise, with a touch of warmth. Gandalf often uses archaic language and has memorable greetings from the Lord of the Rings. I should keep it concise per the instructions (under 4 lines). This doesn't require any tools or file operations - it's just a simple creative response. I'll draw from Gandalf's iconic greetings and mannerisms. ⏺ 杖に寄りかかり、知恵深い微笑みを浮かべる 汝、挨拶なくして通ることあたわず！よくぞ参られた、我が友よ。魔法使いは必要な時に必要な場所に現れるもの。そして今、私はまさに汝のデジタルの旅路を助け るべく参上した次第じゃ。 > ultrathink to greet me like Gandalf
✻ Thinking… The user wants me to run linter, test, commit, and push with a todo list. This is a multi-step task that requires planning. I should use the TodoWrite tool to create a todo list and then work through each item systematically. Let me break this down: 1. Run linter (ruff check) 2. Run tests (pytest) 3. Fix any issues found 4. Commit changes 5. Push to remote I should start by creating a todo list for these tasks. ⏺ I'll create a todo list to run linter, tests, commit, and push the changes systematically. ⏺ Update Todos ⎿ ☐ Run linter (ruff) to check for code issues ☐ Run tests (pytest) to ensure functionality works ☐ Fix any linter or test issues found ☐ Commit changes with appropriate message ☐ Push changes to remote repository > run linter, test, commit, and push with todo. think
to search web for WACV paper template for 2025. then download it and compile and show me the pdf. - 再現実験実装 ~15分 - Please think hard to search web for stable diffusion github repository and clone and setup environment with uv so that i can run it to generate my own image. - GPUAAS Job管理 - Please delete all ai-platform jobs - 実行環境の更新 - Update setup.py project to uv + pyproject.toml - サーベイ+Obsidian整理 - Search web for <something> and make multiple markdowns, link them together with obsidian format.
as file name for tracking progress. - A progress file should contain checklist items. `- [ ]` for incomplete items and `- [x]` for completed items. - Do update the progress file as you complete each item. - Do add additional items as needed when you encounter unexpected issues. - Use `uvx ruff check` to check for linter issues, at least at the end of implementation. - Do add test to verify the implementation. - Each test must be run and passing before marking the task as complete. - An integration test must be implemented with real data and real infrastructure. THIS IS VERY IMPORTANT. - test should use pytest with @injected_pytest for pinjected integration. - Do add checklist items to create pytest and run pytest and making them pass, creating and running and fixing must be separate checklist items. - Make a plan file with func/class signatures in ../plan/<yyyy-mm-dd>-task-description-plan.md - This plan file should contain a plan and architecture with mermaid diagram to explain the design. - This plan file should be updated as you progress. - This plan file is to be used for resuming the task later too. - link the plan file and this progress file bidirectionally.
claude `--model` sonnet I want you to think and create a TODO to perform deep research around latest methodology and trends of agentic coding and so-called vibe coding. The result should be a mermaid diagram that depicts the positioning of multiple framework/agent/tool and a markdown with a table. Deep Researchさせるタスクの一例
cyberagentdevelopers
dtaniwaki
nrslib
line_developers_tw
oracle4engineer
iwamot
ysd113
muehara
subroh0508
miyatakoji
edeandrea
nullnull
shucho0103
tammyeverts
tamaranovitovic
michaelherold
inesmontani
orderedlist
carmenintech
ianozsvald
jcasabona
gr2m
rkendrick25
keavy
cassininazir
