### qzcli Quick Start Commands Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/qzcli/SKILL.md Basic commands to get started with qzcli: login, discover resources, check availability, and list running jobs. ```bash # 1. Login qzcli login ``` ```bash # 2. Discover and cache workspaces/compute groups (run once, re-run after joining new workspaces) qzcli res -u ``` ```bash # 3. Check available nodes qzcli avail ``` ```bash # 4. List running jobs qzcli ls -c -r ``` -------------------------------- ### Install and Authenticate Modal Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/serverless-modal/SKILL.md Install the Modal client and authenticate your account. The `modal setup` command opens a browser for login and saves credentials locally. ```bash pip install modal modal setup # Opens browser login, writes token to ~/.modal.toml # Verify: modal run -q 'print("ok")' ``` -------------------------------- ### Setup Python Virtual Environment and Install Dependencies Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/README.md Use this command to set up a Python virtual environment and install project dependencies from requirements.txt. ```bash python3 -m venv .venv && source .venv/bin/activate pip install -r requirements.txt ``` -------------------------------- ### Modal Setup Command Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/run-experiment/SKILL.md Install the Modal client and set up your account. Ensure a payment method is bound to your Modal account to utilize the free tier. ```bash pip install modal && modal setup ``` -------------------------------- ### Vast.ai Setup Command Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/run-experiment/SKILL.md Install the Vast.ai Python client and set your API key. Upload your SSH public key to Vast.ai to enable instance access. ```bash pip install vastai && vastai set api-key YOUR_KEY ``` -------------------------------- ### Install Dependencies and Setup Codex Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/README.md Install necessary npm packages and set up the Codex CLI for model configuration. Ensure you have Node.js and npm installed. ```bash npm install -g @anthropic-ai/claude-code npm install -g @openai/codex codex setup # set model to gpt-5.4 ``` -------------------------------- ### Vast.ai On-Demand GPU Setup Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/README.md Instructions for setting up Vast.ai for on-demand GPU rentals. Includes installing the Vast.ai CLI, setting API keys, and uploading SSH keys. ```bash pip install vastai ``` ```bash vastai set api-key YOUR_API_KEY ``` ```bash ssh-keygen -t ed25519 -C "your_email@example.com" ``` ```bash cat ~/.ssh/id_ed25519.pub ``` ```bash vastai search offers 'gpu_ram>=24 reliability>0.95' -o 'dph+' --limit 3 ``` -------------------------------- ### Put Context at the Start: Strong Example Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/shared-references/writing-principles.md Demonstrates how to improve sentence flow by stating the problem or context before introducing the new solution. ```text To address the alignment problem, we introduce a new attention mechanism. ``` -------------------------------- ### GEMINI.md Project Instructions Example Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/docs/ANTIGRAVITY_ADAPTATION.md Define project-specific instructions using GEMINI.md format. This includes details for GPU server access, conda environment activation, and code directory setup, as well as research project parameters. ```markdown ## GPU Server (for auto-experiments) - SSH: `ssh my-gpu-server` (key-based auth, no password) - GPU: 4x A100 - Conda env: `research` (Python 3.10 + PyTorch) - Activate: `eval "$(/opt/conda/bin/conda shell.bash hook)" && conda activate research` - Code directory: `/home/user/experiments/` - Use `screen` for background jobs: `screen -dmS exp0 bash -c '...'` ## Research Project - Topic: [your research topic] - Target venue: ICLR/NeurIPS/ICML - Key files: NARRATIVE_REPORT.md, idea-stage/IDEA_REPORT.md ``` -------------------------------- ### Install and Configure Zotero MCP Server Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/README.md Install the zotero-mcp-server using uv or pip. Then, add the Zotero integration to Claude Code, choosing between a local API setup (requires Zotero desktop) or a web API setup using your Zotero API key and User ID. ```bash # Install uv tool install zotero-mcp-server # or: pip install zotero-mcp-server # Add to Claude Code (Local API — requires Zotero desktop running) claude mcp add zotero -s user -- zotero-mcp -e ZOTERO_LOCAL=true # Or use Web API (works without Zotero running) claude mcp add zotero -s user -- zotero-mcp \ -e ZOTERO_API_KEY=your_key -e ZOTERO_USER_ID=your_id ``` -------------------------------- ### Put Context at the Start: Weak Example Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/shared-references/writing-principles.md Presents a sentence where the new material (attention mechanism) is introduced before the problem it solves. ```text A new attention mechanism is introduced to solve the alignment problem. ``` -------------------------------- ### Install qzcli Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/qzcli/SKILL.md Install the qzcli tool and its dependencies using pip and clone the repository. This is a prerequisite for using the tool. ```bash pip install rich requests prompt_toolkit mcp git clone https://github.com/tianyilt/qzcli_tool cd qzcli_tool && pip install -e . ``` -------------------------------- ### Idea Discovery Pipeline Command Example Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/idea-discovery/SKILL.md Example of how to invoke the idea discovery skill with custom parameters. ```bash /idea-discovery "topic" — ref paper: https://arxiv.org/abs/2406.04329 ``` ```bash /idea-discovery "topic" — compact: true ``` -------------------------------- ### Install exa-py SDK Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/exa-search/SKILL.md Install the exa-py SDK using pip. This is a prerequisite for using the Exa search tool. ```bash pip install exa-py ``` -------------------------------- ### Display Effort Configuration Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/shared-references/effort-contract.md Skills should print their current effort configuration at the start for transparency. This example shows the format including effort level and adjusted constants. ```text ⚡ [effort: max] papers=25, ideas=16, rounds=6 | Codex: xhigh (always) ``` -------------------------------- ### Example Prompt for Image Generation Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/mcp-servers/codex-image2/README.md An example text prompt to be used with the `generate_start` tool for image generation. ```text Generate a clean blue circle on white background. ``` -------------------------------- ### Install and Run PPTX Generation Script Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/skills-codex-gemini-review/paper-poster/SKILL.md Installs the python-pptx library if not present and then executes the poster generation script. ```bash python3 -c "import pptx" 2>/dev/null || pip install python-pptx cd poster && python3 generate_pptx.py # Output: poster/poster.pptx ``` -------------------------------- ### Install DeepXiv SDK Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/deepxiv/SKILL.md Install the DeepXiv Python SDK using pip. This is required if the deepxiv CLI is not already installed. ```bash pip install deepxiv-sdk ``` -------------------------------- ### Install python-pptx Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/paper-slides/SKILL.md Installs the python-pptx library if not already present. This is required for generating PowerPoint presentations. ```bash python3 -c "import pptx" 2>/dev/null || pip install python-pptx ``` -------------------------------- ### Create Virtual Environment and Install Dependencies for llm-chat Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/docs/TRAE_ARIS_RUNBOOK_EN.md Set up a Python virtual environment and install the necessary dependencies for the llm-chat MCP server. ```bash cd D:\path\to\Auto-claude-code-research-in-sleep python -m venv .venv .\.venv\Scripts\pip install -r mcp-servers\llm-chat\requirements.txt ``` -------------------------------- ### Install TeX Live from Source (User Directory) Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/skills-codex-gemini-review/paper-poster/SKILL.md Install TeX Live directly from the CTAN mirror into your user directory. This method does not require `sudo` and is reliable in non-interactive environments. Remember to replace 'YYYY' with the current year. ```bash # Option 3: User-directory install (NO sudo needed — always works) curl -L https://mirror.ctan.org/systems/texlive/tlnet/install-tl-unx.tar.gz | tar xz cd install-tl-* cat > texlive.profile << 'PROF' selected_scheme scheme-basic TEXDIR ~/texlive/YYYY TEXMFLOCAL ~/texlive/texmf-local TEXMFSYSCONFIG ~/texlive/YYYY/texmf-config TEXMFSYSVAR ~/texlive/YYYY/texmf-var TEXMFHOME ~/texmf binary_x86_64-darwin 1 instopt_adjustpath 0 instopt_adjustrepo 1 instopt_write18_restricted 1 tlpdbopt_autobackup 1 tlpdbopt_install_docfiles 0 tlpdbopt_install_srcfiles 0 PROF ./install-tl --profile=texlive.profile export PATH="$HOME/texlive/YYYY/bin/universal-darwin:$PATH" ``` -------------------------------- ### Install and Configure Oracle for Claude Code Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/README.md These bash commands guide the installation of the Oracle tool and its integration with Claude Code. It covers adding the Oracle MCP and setting up API keys for enhanced reviewer capabilities. ```bash # 1. Install Oracle npm install -g @steipete/oracle # 2. Add Oracle MCP to Claude Code claude mcp add oracle -s user -- oracle-mcp # 3. Restart Claude Code # 4a. API mode (fast, recommended): export OPENAI_API_KEY="your-key" # 4b. Browser mode (free, no API key — log in to ChatGPT in Chrome): ``` -------------------------------- ### Invention Disclosure Structure Example Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/invention-structuring/SKILL.md This example illustrates the hierarchical structure of claims, starting from an independent method claim and branching into dependent claims that narrow down features or add optional elements. Use this as a template for mapping claim dependencies. ```plaintext Independent Claim 1 (method, broadest scope) ├── Core inventive feature A ├── Core inventive feature B └── Known feature C (for context) Dependent Claim 2 → narrows feature A with specific implementation Dependent Claim 3 → narrows feature B with specific parameters Dependent Claim 4 → depends on 2, adds optional feature D Dependent Claim 5 → alternative implementation of feature A ``` -------------------------------- ### Overleaf Setup Instruction Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/overleaf-sync/SKILL.md This is the instruction printed to the user for setting up the Overleaf sync. It must be run in the user's terminal, not through the agent. ```bash Run this in your own terminal (NOT through me): bash /tools/overleaf_setup.sh When it finishes, tell me "setup done" and I'll verify. ``` -------------------------------- ### Async Review Job Status Response Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/mcp-servers/claude-review/README.md Example response after starting an asynchronous review job, indicating the job ID and its initial status. ```json { "jobId": "5d8d0a9c5a2f4f42ae44f6f0c2d73f6f", "status": "queued", "done": false } ``` -------------------------------- ### Start Codex in Project Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/docs/CODEX_CLAUDE_REVIEW_GUIDE.md Initiate Codex within your project directory to begin using its features. ```bash codex -C /path/to/your/project ``` -------------------------------- ### Vast GPU Skill Workflow Example Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/vast-gpu/SKILL.md Illustrates the typical workflow when composing the 'train model' experiment with the Vast GPU skill. This includes automatic GPU detection, provisioning, user interaction for offer selection, and automatic destruction upon completion. ```bash /run-experiment "train model" ← detects gpu: vast, calls /vast-gpu provision ↳ /vast-gpu provision ← analyzes task, presents options with cost ↳ user picks option ← rent + setup + deploy ↳ /vast-gpu destroy ← auto-destroy when done (if auto_destroy: true) ``` -------------------------------- ### Poster LaTeX Document Setup Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/skills-codex/paper-poster/SKILL.md Sets up the document class, essential packages for geometry and tcbposter, and defines custom colors for the poster theme. Ensure all specified packages are installed. ```latex \documentclass{article} \usepackage[paperwidth=1189mm,paperheight=841mm,margin=0mm]{geometry} \usepackage{tcolorbox} \tcbuselibrary{poster,skins,fitting} \usepackage{graphicx} \usepackage{amsmath,amssymb} \usepackage{enumitem} \usepackage[table]{xcolor} \usepackage{lmodern} \usepackage[T1]{fontenc} \pagestyle{empty} % ── Venue Color Theme ── \definecolor{primary}{HTML}{VENUE_PRIMARY} % deep, saturated \definecolor{secondary}{HTML}{VENUE_SECONDARY} % medium \definecolor{accent}{HTML}{VENUE_ACCENT} % contrast \definecolor{bgposter}{HTML}{VENUE_BG_DEEP} % poster background (NOT white, use tinted) \definecolor{redbg}{HTML}{FFF5F3} % card backgrounds (tinted, NOT white) \definecolor{bluebg}{HTML}{F0F4FF} \definecolor{darkbg}{HTML}{FDF6F3} \definecolor{redtitlebg}{HTML}{FDEAE8} % card title bar backgrounds \definecolor{bluetitlebg}{HTML}{E4ECFF} \definecolor{darktitlebg}{HTML}{F5E8E2} \definecolor{textdark}{HTML}{111827} \definecolor{textgray}{HTML}{4B5563} \definecolor{stathighlight}{HTML}{FEE8E8} \pagecolor{bgposter} \color{textdark} % ── List styling ── \setlist[itemize]{leftmargin=24pt, itemsep=6pt, parsep=2pt, topsep=2pt, label={\color{secondary}$\blacktriangleright$}} \setlist[enumerate]{leftmargin=24pt, itemsep=6pt, parsep=2pt, topsep=2pt, label={\color{primary}\bfseries\arabic*.}} % ── Figure+caption macro (ensures uniform spacing) ── \newcommand{\posterfig}[3]{ \centering\includegraphics[width=#1\linewidth]{#2}\\[3mm] {\fontsize{26}{32}\selectfont\color{textgray}\textit{#3}}\vspace{2mm}% } % ── Card styles (left accent stripe design) ── \tcbset{ redcard/.style={ enhanced, arc=0pt, boxrule=0pt, colback=redbg, borderline west={5pt}{0pt}{secondary}, left=16pt, right=14pt, top=4pt, bottom=4pt, fonttitle=\fontsize{40}{48}\selectfont\bfseries\color{secondary}, coltitle=secondary, colbacktitle=redtitlebg, toptitle=6pt, bottomtitle=6pt, titlerule=2pt, titlerule style={secondary!50}, valign=top, drop shadow={opacity=0.18}, }, bluecard/.style={ enhanced, arc=0pt, boxrule=0pt, colback=bluebg, borderline west={5pt}{0pt}{accent}, left=16pt, right=14pt, top=4pt, bottom=4pt, fonttitle=\fontsize{40}{48}\selectfont\bfseries\color{accent}, coltitle=accent, colbacktitle=bluetitlebg, toptitle=6pt, bottomtitle=6pt, titlerule=2pt, titlerule style={accent!50}, valign=top, drop shadow={opacity=0.18}, }, darkcard/.style={ enhanced, arc=0pt, boxrule=0pt, colback=darkbg, borderline west={5pt}{0pt}{primary}, left=16pt, right=14pt, top=4pt, bottom=4pt, fonttitle=\fontsize{40}{48}\selectfont\bfseries\color{primary}, coltitle=primary, colbacktitle=darktitlebg, toptitle=6pt, bottomtitle=6pt, titlerule=2pt, titlerule style={primary!50}, valign=top, drop shadow={opacity=0.18}, }, highlightcard/.style={ enhanced, arc=0pt, boxrule=0pt, colback=primary!18!white, borderline west={6pt}{0pt}{primary}, left=16pt, right=14pt, top=4pt, bottom=4pt, fonttitle=\fontsize{40}{48}\selectfont\bfseries\color{white}, coltitle=white, colbacktitle=primary, toptitle=6pt, bottomtitle=6pt, valign=top, drop shadow={opacity=0.22}, }, } \begin{document} \begin{tcbposter}[ coverage={spread}, poster={columns=4, rows=20, spacing=0mm}, % Use columns=3 for portrait A0 ] % ══ TITLE BAR ══ \posterbox[ enhanced, colback=primary, colframe=primary, colupper=white, arc=0pt, boxrule=0pt, left=40pt, right=40pt, top=12pt, bottom=8pt, halign=center, valign=center, drop shadow={opacity=0.3} ]{name=title, column=1, span=4, between=top and row4}{ {\fontsize{84}{100}\selectfont\bfseries PAPER TITLE}\\[12pt] {\fontsize{36}{44}\selectfont Authors}\\[8pt] {\fontsize{30}{38}\selectfont\color{white!70} Affiliations | VENUE YEAR | github.com/...} } % ══ STATS BANNER ══ \posterbox[ enhanced, colback=primary!15!white, boxrule=0pt, arc=0pt, left=12pt, right=12pt, top=6pt, bottom=6pt, valign=center, borderline south={3pt}{0pt}{primary!35}, ]{name=stats, column=1, span=4, between=row4 and row6}{ \centering \begin{minipage}[c]{0.235\linewidth}\centering \fcolorbox{primary!40}{stathighlight}{\parbox{0.88\linewidth}{ \centering\vspace{6pt}% {\fontsize{66}{80}\selectfont\bfseries\color{primary} STAT1}\\[4pt] {\fontsize{26}{32}\selectfont\color{textdark} Label 1}\vspace{6pt}% }} \end{minipage}\hfill % ... 3 more stat callouts in same pattern } % ══ CONTENT CARDS ══ % Use card styles: \posterbox[redcard, title={...}]{...}{...} % Body text: \fontsize{34}{44}\selectfont % Figures: \posterfig{0.96}{figures/name.png}{Caption.} % Colorboxes: \colorbox{primary!20}{\parbox{0.94\linewidth}{...}} \end{tcbposter} \end{document} ``` -------------------------------- ### Initialize Research Wiki Source: https://context7.com/wanshuiyin/auto-claude-code-research-in-sleep/llms.txt Initializes a new research wiki directory for a project. This command should be run first to set up the necessary file structure. ```bash python3 tools/research_wiki.py init research-wiki/ ``` -------------------------------- ### Configure qzcli credentials Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/qzcli/SKILL.md Set up authentication for qzcli. The recommended method is using an environment file. Alternatively, environment variables can be used. ```bash # Option A: env file (recommended) mkdir -p ~/.qzcli cat > ~/.qzcli/.env <<'EOF' QZCLI_USERNAME="your_username" QZCLI_PASSWORD="your_password" EOF ``` ```bash # Option B: environment variables export QZCLI_USERNAME="your_username" export QZCLI_PASSWORD="your_password" export QZCLI_API_URL="https://qz.yourorg.edu.cn" ``` -------------------------------- ### Verify PyTorch and CUDA Setup on Vast.ai Instance Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/vast-gpu/SKILL.md Verifies the PyTorch installation, CUDA availability, and the number of detected GPUs on a remote Vast.ai instance. This command is executed after code synchronization. ```bash ssh -p root@ "cd /workspace/project && python -c 'import torch; print(f"PyTorch {torch.__version__}, CUDA: {torch.cuda.is_available()}, GPUs: {torch.cuda.device_count()}")'" ``` -------------------------------- ### Start Gemini Review Prompt Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/skills-codex-gemini-review/research-refine/SKILL.md This prompt guides Gemini to act as a senior ML reviewer, focusing on method-first research proposals. It specifies review principles, scoring dimensions, and output format for feedback. ```yaml mcp__gemini-review__review_start: prompt: | You are a senior ML reviewer for a top venue (NeurIPS/ICML/ICLR). This is an early-stage, method-first research proposal. Your job is NOT to reward extra modules, contribution sprawl, or a giant benchmark checklist. Your job IS to stress-test whether the proposed method: (1) still solves the original anchored problem, (2) is concrete enough to implement, (3) presents a focused, elegant contribution, (4) uses foundation-model-era techniques appropriately when they are the natural fit. Review principles: - Prefer the smallest adequate mechanism over a larger system. - Penalize parallel contributions that make the paper feel unfocused. - If a modern LLM / VLM / Diffusion / RL route would clearly produce a better paper, say so concretely. - If the proposal is already modern enough, do NOT force trendy components. - Do not ask for extra experiments unless they are needed to prove the core claims. Read the Problem Anchor first. If your suggested fix would change the problem being solved, call that out explicitly as drift instead of treating it as a normal revision request. === PROPOSAL === [Paste the FULL proposal from Phase 1] === END PROPOSAL === Score these 7 dimensions from 1-10: 1. **Problem Fidelity**: Does the method still attack the original bottleneck, or has it drifted into solving something easier or different? 2. **Method Specificity**: Are the interfaces, representations, losses, training stages, and inference path concrete enough that an engineer could start implementing? 3. **Contribution Quality**: Is there one dominant mechanism-level contribution with real novelty, good parsimony, and no obvious contribution sprawl? 4. **Frontier Leverage**: Does the proposal use current foundation-model-era primitives appropriately when they are the right tool, instead of defaulting to old-school module stacking? 5. **Feasibility**: Can this method be trained and integrated with the stated resources and data assumptions? 6. **Validation Focus**: Are the proposed experiments minimal but sufficient to validate the core claims? Is there unnecessary experimental bloat? 7. **Venue Readiness**: If executed well, would the contribution feel sharp and timely enough for a top venue? **OVERALL SCORE** (1-10): Weighted toward Problem Fidelity, Method Specificity, Contribution Quality, and Frontier Leverage. Use this weighting: Problem Fidelity 15%, Method Specificity 25%, Contribution Quality 25%, Frontier Leverage 15%, Feasibility 10%, Validation Focus 5%, Venue Readiness 5%. For each dimension scoring < 7, provide: - The specific weakness - A concrete fix at the method level (interface / loss / training recipe / integration point / deletion of unnecessary parts) - Priority: CRITICAL / IMPORTANT / MINOR Then add: - **Simplification Opportunities**: 1-3 concrete ways to delete, merge, or reuse components while preserving the main claim. Write "NONE" if already tight. - **Modernization Opportunities**: 1-3 concrete ways to replace old-school pieces with more natural foundation-model-era primitives if genuinely better. Write "NONE" if already modern enough. - **Drift Warning**: "NONE" if the proposal still solves the anchored problem; otherwise explain the drift clearly. - **Verdict**: READY / REVISE / RETHINK Verdict rule: - READY: overall score >= 9, no meaningful drift, one focused dominant contribution, and no obvious complexity bloat remains - REVISE: the direction is promising but not yet at READY bar - RETHINK: the core mechanism or framing is still fundamentally off ``` -------------------------------- ### Monitor Experiment Queue Status Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/tools/experiment_queue/README.md Queries the scheduler's state file (`/tmp/queue_state.json`) on the remote server to get a count of jobs by their current status (completed, running, pending). Requires `jq` to be installed on the server. ```bash ssh 'jq ".jobs | group_by(.status) | map({(.[0].status): length}) | add" /tmp/queue_state.json' ``` -------------------------------- ### Install Exa AI SDK and Set API Key Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/README.md Install the Exa Python SDK for AI-powered web search and set your API key. This enables searching across blogs, documentation, news, and research papers with content extraction. ```bash pip install exa-py export EXA_API_KEY=your-key-here ``` -------------------------------- ### Grant Proposal Command Example Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/grant-proposal/SKILL.md Demonstrates how to invoke the grant proposal skill with specific arguments for topic, grant type, and output format. ```bash /grant-proposal "topic — NSF CAREER, latex output" ``` -------------------------------- ### Execute Exa Search with Filters Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/exa-search/SKILL.md Execute an Exa search with specific filters, including category, start date, and content extraction settings. This example searches for 'QUERY' as a research paper published after 2025-01-01, extracting full text up to 8000 characters. ```bash python3 "$SCRIPT" search "QUERY" --max 10 \ --category "research paper" \ --start-date 2025-01-01 \ --content text --max-chars 8000 ``` -------------------------------- ### Start Claude Review with Specific Prompt Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/skills-codex-claude-review/research-refine/SKILL.md Initiates a review process by sending a detailed prompt to Claude. The prompt guides the AI to act as a senior ML reviewer, focusing on specific criteria like problem fidelity, method specificity, and contribution quality. Save the returned jobId for status polling. ```yaml mcp__claude-review__review_start: prompt: | You are a senior ML reviewer for a top venue (NeurIPS/ICML/ICLR). This is an early-stage, method-first research proposal. Your job is NOT to reward extra modules, contribution sprawl, or a giant benchmark checklist. Your job IS to stress-test whether the proposed method: (1) still solves the original anchored problem, (2) is concrete enough to implement, (3) presents a focused, elegant contribution, (4) uses foundation-model-era techniques appropriately when they are the natural fit. Review principles: - Prefer the smallest adequate mechanism over a larger system. - Penalize parallel contributions that make the paper feel unfocused. - If a modern LLM / VLM / Diffusion / RL route would clearly produce a better paper, say so concretely. - If the proposal is already modern enough, do NOT force trendy components. - Do not ask for extra experiments unless they are needed to prove the core claims. Read the Problem Anchor first. If your suggested fix would change the problem being solved, call that out explicitly as drift instead of treating it as a normal revision request. === PROPOSAL === [Paste the FULL proposal from Phase 1] === END PROPOSAL === Score these 7 dimensions from 1-10: 1. **Problem Fidelity**: Does the method still attack the original bottleneck, or has it drifted into solving something easier or different? 2. **Method Specificity**: Are the interfaces, representations, losses, training stages, and inference path concrete enough that an engineer could start implementing? 3. **Contribution Quality**: Is there one dominant mechanism-level contribution with real novelty, good parsimony, and no obvious contribution sprawl? 4. **Frontier Leverage**: Does the proposal use current foundation-model-era primitives appropriately when they are the right tool, instead of defaulting to old-school module stacking? 5. **Feasibility**: Can this method be trained and integrated with the stated resources and data assumptions? 6. **Validation Focus**: Are the proposed experiments minimal but sufficient to validate the core claims? Is there unnecessary experimental bloat? 7. **Venue Readiness**: If executed well, would the contribution feel sharp and timely enough for a top venue? **OVERALL SCORE** (1-10): Weighted toward Problem Fidelity, Method Specificity, Contribution Quality, and Frontier Leverage. Use this weighting: Problem Fidelity 15%, Method Specificity 25%, Contribution Quality 25%, Frontier Leverage 15%, Feasibility 10%, Validation Focus 5%, Venue Readiness 5%. For each dimension scoring < 7, provide: - The specific weakness - A concrete fix at the method level (interface / loss / training recipe / integration point / deletion of unnecessary parts) - Priority: CRITICAL / IMPORTANT / MINOR Then add: - **Simplification Opportunities**: 1-3 concrete ways to delete, merge, or reuse components while preserving the main claim. Write "NONE" if already tight. - **Modernization Opportunities**: 1-3 concrete ways to replace old-school pieces with more natural foundation-model-era primitives if genuinely better. Write "NONE" if already modern enough. - **Drift Warning**: "NONE" if the proposal still solves the anchored problem; otherwise explain the drift clearly. - **Verdict**: READY / REVISE / RETHINK Verdict rule: - READY: overall score >= 9, no meaningful drift, one focused dominant contribution, and no obvious complexity bloat remains - REVISE: the direction is promising but not yet at READY bar - RETHINK: the core mechanism or framing is still fundamentally off ``` -------------------------------- ### Clone Repository and Install Skills Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/README.md Clone the project repository, navigate into the directory, copy skills to the .claude/skills directory, and run the main claude command. This step is crucial for setting up the project locally. ```bash git clone https://github.com/wanshuiyin/Auto-claude-code-research-in-sleep.git cd Auto-claude-code-research-in-sleep cp -r skills/* ~/.claude/skills/ claude ``` -------------------------------- ### Install Claude-Review Overrides Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/skills-codex-claude-review/README.md Installs the Claude-review override skills by copying them to the local Codex skills directory. This should be done after installing the base Codex skills. ```bash cp -a skills/skills-codex-claude-review/* ~/.codex/skills/ ``` -------------------------------- ### Set up llm-chat MCP Server with Virtual Environment Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/docs/CURSOR_ADAPTATION.md Set up an alternative reviewer using the llm-chat MCP server. This requires creating a Python virtual environment and installing dependencies. ```bash cd /path/to/Auto-claude-code-research-in-sleep python3 -m venv .venv .venv/bin/pip install -r mcp-servers/llm-chat/requirements.txt ``` -------------------------------- ### Install PyMuPDF for SVG Conversion Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/paper-poster/SKILL.md Installs the PyMuPDF library, which is the preferred method for converting PDF files to SVG. It's installed using pip. ```bash python3 -c "import fitz" 2>/dev/null || pip install pymupdf ``` -------------------------------- ### Setup Meta-Optimize Workflow Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/README.md Run this one-time setup in your terminal to prepare the necessary directories, copy configuration and script files, and make scripts executable. The Claude hooks become active immediately after running 'claude'. ```bash mkdir -p .claude .aris/meta tools/meta_opt cp Auto-claude-code-research-in-sleep/templates/claude-hooks/meta_logging.json .claude/settings.json cp Auto-claude-code-research-in-sleep/tools/meta_opt/*.sh tools/meta_opt/ chmod +x tools/meta_opt/*.sh claude # hooks active immediately ``` -------------------------------- ### Deploy Local Experiment (Bash/Zsh) Source: https://github.com/wanshuiyin/auto-claude-code-research-in-sleep/blob/main/skills/skills-codex/run-experiment/SKILL.md Run a local experiment, specifying the visible GPU for CUDA or letting PyTorch automatically use MPS on Mac. Logs are directed to a file. ```bash # Linux with CUDA CUDA_VISIBLE_DEVICES= python