wanshuiyin/Auto-claude-code-research-in-sleep

Ablation Planner designs ablation studies for machine learning research papers, isolating component contributions and prioritizing experiments when main results pass.

AlphaXiv returns LLM-optimized summaries for a single paper from AlphaXiv, with tiered source fallback. Use when asked to explain or summarize a paper, or given an arXiv/AlphaXiv URL or ID.

Analyze Results analyzes ML experiment results, computes statistics, and generates comparison tables and insights. Use when the user says "analyze results", "compare", or needs to interpret experimental data.

ArXiv Paper Search searches arXiv for papers, fetches details by ID, downloads PDFs, and summarizes them; use it when the user asks to search arxiv, get a paper, download a paper, or save PDFs to a local library.

Auto Paper Improvement Loop iteratively improves a generated paper via external LLM review, fix implementation, and recompilation for two rounds. Use it after paper generation to catch structural and presentation issues.

Auto Review Loop autonomously iterates review, implements fixes, and re-reviews until an independent reviewer gives a policy-approved positive assessment or MAX_ROUNDS is reached.

Auto Review Loop LLM runs an autonomous review-improve-re-review loop for research, using any OpenAI-compatible LLM API as an external reviewer until it gives a positive assessment or max rounds are reached. Configure via llm-chat MCP server or environment variables.

Auto Review Loop MiniMax runs an autonomous multi-round research review loop that calls the MiniMax API for external reviewer feedback, then implements fixes and re-reviews until a positive verdict or max rounds. Use it when you want MiniMax instead of Codex MCP for external review.

Citation Audit verifies every bibliographic entry in a paper is real, correctly attributed, and cited in a context the cited paper actually supports, catching hallucinated authors, wrong years, fabricated venues, version mismatches, and wrong-context citations. Use it when checking citations, verifying references, or before submission to ensure bibliography integrity.

Claims Drafting drafts patent claims for an invention based on a disclosure, prior art report, and novelty assessment. Use it when writing claims in US, EP, or CN jurisdictions, including utility model apparatus-only claims.

Communications Literature Review finds and synthesizes academic papers on wireless, networking, and communication systems using a knowledge-base-first retrieval workflow. Use it when you need related work, a survey, or a landscape summary for a communications research topic.

DSE Loop autonomously explores design spaces for computer architecture and EDA by running a program, analyzing results, tuning parameters, and iterating until an objective is met or a timeout occurs. Use it when the user asks for DSE, parameter sweeps, optimization, or finding the best configuration.

DeepXiv searches and progressively reads open-access academic papers through the DeepXiv SDK. Use when you need layered paper access, section-level reading, trending papers, or DeepXiv-backed literature retrieval.

Embodiment Description writes detailed patent specification embodiments that teach how to make and use an invention. Use it when drafting patent applications or when asked to write an embodiment, detailed description, or 实施例描述.

Exa Search searches the web via Exa's API and extracts content like highlights, full text, or summaries alongside results. Use it when you need broad web results beyond academic databases, find similar pages, or fetch content from known URLs.

Experiment Audit verifies experiment integrity before results are claimed, using cross-model review to check for fake ground truth, score normalization fraud, phantom results, and insufficient scope. Use it when the user asks to audit experiments or after experiments complete but before writing claims.

Experiment Bridge turns an experiment plan into running GPU experiments by implementing code, reviewing it, deploying it, and collecting initial results. Use it after idea refinement and before the auto review loop.

Experiment Plan turns a refined research proposal into a detailed, claim-driven experiment roadmap. It specifies experiment blocks, baselines, metrics, and run order to defend a paper's core claims.

Experiment Queue orchestrates large batches of ML experiments on SSH remote GPU servers with state tracking, OOM retry, stale-screen cleanup, and wave transitions. Use it for multi-seed sweeps, grid searches, teacher-student chains, or any 10+ job batch that needs orchestration beyond a single run.

Feishu Notification sends status updates and checkpoint alerts to Feishu/Lark via webhook or interactive bridge. Use it when a user asks to notify Feishu or when other skills need to report experiment results, review scores, or pipeline steps.

Figure Description processes user-provided patent figures and generates formal drawing descriptions with reference numerals. Use it when you need to create figure captions, assign reference numerals, and cross-reference them to patent claims.

FigureSpec generates deterministic publication-quality architecture, workflow, and pipeline diagrams from structured JSON into editable SVG. Use it when you need precise, editable, publication-ready vector figures for architecture or workflow diagrams rather than AI illustration.

Formula Derivation structures and derives research formulas, turning scattered equations and assumptions into a coherent derivation package or a blocker report. Use it when the target formula is not yet fixed or when you need a theory line rather than a finished theorem proof.

Gemini Search searches research papers via Gemini for broad literature discovery, decomposing topics into sub-problems and naming variants to surface papers traditional API searches may miss. Use it when a user says "gemini search", "gemini papers", or wants AI-powered research discovery beyond arXiv and Semantic Scholar.

Grant Proposal drafts structured, reviewer-ready grant applications from research ideas and literature, supporting KAKENHI, NSF, NSFC, ERC, DFG, SNSF, ARC, NWO, and generic formats. Use it when turning research ideas into a funding application.