debuffer-skills

Workflow 1: Full idea discovery pipeline. Orchestrates research-lit → idea-creator → novelty-check → research-review → research-refine plus experiment-plan to go from a broad research direction to validated, pilot-tested ideas. Also handles robotics / embodied AI directions through simulation-first robotics mode. Use when user says "找idea全流程", "idea discovery pipeline", "从零开始找方向", "robotics idea discovery", "embodied AI idea", or wants the complete idea exploration workflow.

Generate and rank research ideas given a broad direction. Use when user says "找idea", "brainstorm ideas", "generate research ideas", "what can we work on", or wants to explore a research area for publishable directions.

Turn one core reference paper and up to three comparable papers into evidence-grounded topic candidates. Use when the user asks to 拆解对标论文, 从论文找选题, analyze reference papers for research ideas, map conflicting findings, or wants a reference-led idea start before novelty checking.

Turn a vague research direction into a problem-anchored, elegant, frontier-aware, implementation-oriented method plan via iterative GPT-5.4 review. Use when the user says "refine my approach", "帮我细化方案", "decompose this problem", "打磨idea", "refine research plan", "细化研究方案", or wants a concrete research method that stays simple, focused, and top-venue ready instead of a vague or overbuilt idea.

Generate a structured paper outline from audited formal experiment results and review conclusions. Use when user says "写大纲", "paper outline", "plan the paper", "论文规划", or wants a paper plan after formal runs and evidence audit. If only smoke, pilot, toy, or validation results exist, produce a gap report or next actions instead of routing to manuscript drafting.

Draft LaTeX paper section by section from an outline. Use when user says "写论文", "write paper", "draft LaTeX", "开始写", or wants to generate LaTeX content from a paper plan.

Workflow 3: evidence-gated paper writing pipeline. Requires audited formal experiment evidence and an accepted paper plan before manuscript drafting. Orchestrates paper-plan → paper-figure → figure-spec/paper-illustration/mermaid-diagram → paper-write → paper-compile → auto-paper-improvement-loop only after the manuscript entry gate passes. If only smoke, pilot, toy, or validation results exist, stop and produce next actions or a gap report. Use when user says "写论文全流程", "write paper pipeline", "从报告到PDF", "paper writing", or wants complete paper generation from audited evidence.

Prompt-only by default in the lightweight pack; Claude/Gemini reviewer transport is explicit opt-in. Generate and rank research ideas given a broad direction. Use when user says "找idea", "brainstorm ideas", "generate research ideas", "what can we work on", or wants to explore a research area for publishable directions.

Prompt-only by default in the lightweight pack; Claude/Gemini reviewer transport is explicit opt-in. Workflow 1: Full idea discovery pipeline. Orchestrates research-lit → idea-creator → novelty-check → research-review to go from a broad research direction to validated, pilot-tested ideas. Use when user says \"找idea全流程\", \"idea discovery pipeline\", \"从零开始找方向\", or wants the complete idea exploration workflow.

Workflow 1: Full idea discovery pipeline. Orchestrates research-lit → idea-creator → novelty-check → research-review → research-refine plus experiment-plan to go from a broad research direction to validated, pilot-tested ideas. Also handles robotics / embodied AI directions through simulation-first robotics mode. Use when user says \"找idea全流程\", \"idea discovery pipeline\", \"从零开始找方向\", \"robotics idea discovery\", \"embodied AI idea\", or wants the complete idea exploration workflow.

Generate and rank research ideas given a broad direction. Use when user says "找idea", "brainstorm ideas", "generate research ideas", "what can we work on", or wants to explore a research area for publishable directions.

Turn one core reference paper and up to three comparable papers into evidence-grounded topic candidates. Use when the user asks to 拆解对标论文, 从论文找选题, analyze reference papers for research ideas, map conflicting findings, or wants a reference-led idea start before novelty checking.

Turn a vague research direction into a problem-anchored, elegant, frontier-aware, implementation-oriented method plan via iterative GPT-5.4 review. Use when the user says "refine my approach", "帮我细化方案", "decompose this problem", "打磨idea", "refine research plan", "细化研究方案", or wants a concrete research method that stays simple, focused, and top-venue ready instead of a vague or overbuilt idea.

Generate a structured paper outline from audited formal experiment results and review conclusions. Use when user says \"写大纲\", \"paper outline\", \"plan the paper\", \"论文规划\", or wants a paper plan after formal runs and evidence audit. If only smoke, pilot, toy, or validation results exist, produce a gap report or next actions instead of routing to manuscript drafting.

Draft LaTeX paper section by section from an outline. Use when user says \"写论文\", \"write paper\", \"draft LaTeX\", \"开始写\", or wants to generate LaTeX content from a paper plan.

Workflow 3: evidence-gated paper writing pipeline. Requires audited formal experiment evidence and an accepted paper plan before manuscript drafting. Orchestrates paper-plan → paper-figure → figure-spec/paper-illustration/mermaid-diagram → paper-write → paper-compile → auto-paper-improvement-loop only after the manuscript entry gate passes. If only smoke, pilot, toy, or validation results exist, stop and produce next actions or a gap report. Use when user says "写论文全流程", "write paper pipeline", "从报告到PDF", "paper writing", or wants complete paper generation from audited evidence.

Turn a refined research proposal or method idea into a detailed, question- and mechanism-driven experiment roadmap. Use after `research-refine`, or when the user asks for a detailed experiment plan, ablation matrix, evaluation protocol, run order, compute budget, or paper-ready validation that supports the core problem, novelty, simplicity, and any LLM / VLM / Diffusion / RL-based contribution.

Revise and polish academic drafts into clear, direct, human-sounding prose while preserving claims, numbers, citations, limitations, task type, and recommendation stance. Implicitly use when the user says 改写, 改稿, 润色, 润色论文, 像人一点, 更像人写的, 不要AI味, 减少AI味, 不要防御性写作, 去掉防御性表达, 写得自然一点, humanize, less defensive, reduce generic AI-style prose, strengthen an abstract or section, or requests a ResNet, Transformer, BERT, Raft, or MapReduce tone.

Apply anti-defensive editing and curated classic-paper structures to revise academic prose into clear, direct, human-sounding writing while preserving claims, numbers, citations, limitations, task type, and recommendation stance. Use when a higher-level skill routes revision, polishing, or humanization work here.

Lightweight AutoDL-first research pipeline: idea discovery → blueprint → experiment planning/AutoDL readiness checks → prompt-only review/audit → optional evidence-checked paper planning. Adapts to venue-only, reference-paper/codebase, idea-doc, existing-repo, or partial-results starts; avoids heavy local compute, defaults to concise artifacts, and prepares AutoDL/HPC approval-bounded runs. Use when user says "全流程", "full pipeline", "从找idea到投稿", "end-to-end research", or wants a complete but user-gated research lifecycle. Manuscript drafting is allowed only after formal runs and evidence audit pass.

Turn a refined research proposal or method idea into a detailed, question- and mechanism-driven experiment roadmap. Use after `research-refine`, or when the user asks for a detailed experiment plan, ablation matrix, evaluation protocol, run order, compute budget, or paper-ready validation that supports the core problem, novelty, simplicity, and any LLM / VLM / Diffusion / RL-based contribution.

Revise and polish academic drafts into clear, direct, human-sounding prose while preserving claims, numbers, citations, limitations, task type, and recommendation stance. Implicitly use when the user says 改写, 改稿, 润色, 润色论文, 像人一点, 更像人写的, 不要AI味, 减少AI味, 不要防御性写作, 去掉防御性表达, 写得自然一点, humanize, less defensive, reduce generic AI-style prose, strengthen an abstract or section, or requests a ResNet, Transformer, BERT, Raft, or MapReduce tone.

Lightweight AutoDL-first research pipeline: idea discovery → blueprint → experiment planning/AutoDL readiness checks → prompt-only review/audit → optional evidence-checked paper planning. Adapts to venue-only, reference-paper/codebase, idea-doc, existing-repo, or partial-results starts; avoids heavy local compute, defaults to concise artifacts, and prepares AutoDL/HPC approval-bounded runs. Use when user says "全流程", "full pipeline", "从找idea到投稿", "end-to-end research", or wants a complete but user-gated research lifecycle. Manuscript drafting is allowed only after formal runs and evidence audit pass.

Prepare, validate, and gate AutoDL/HPC research experiments with GitHub deploy-key bootstrap, offline data policy, preflight and smoke gates, FileZilla/SFTP result transfer, and formal-run approval boundaries. Defaults to command preparation over autonomous SSH execution. Use when Codex needs AutoDL, /root/autodl-tmp, remote GPU/HPC smoke tests, preflight_autodl.py, run_autodl_smoke.sh, deploy keys, formal suite gating, or safe download/audit workflows.

Opt-in SSH job queue for multi-seed/multi-config ML experiments with OOM-aware retry, stale-screen cleanup, and wave-transition race prevention. In the lightweight AutoDL-first pack, use only when the user explicitly asks for SSH queue orchestration or approves a prepared command block.

Prepare and gate ML experiment execution with AutoDL/HPC preferred for heavy compute. Runs only local validation/tiny smoke checks by default, prints SSH/remote command blocks for approval, and keeps Vast.ai/Modal as opt-in alternatives. Use when user says "run experiment", "deploy to server", "跑实验", or needs to launch training jobs.

Persistent research knowledge base that accumulates papers, ideas, experiments, claims, and their relationships across the entire research lifecycle. Inspired by Karpathy's LLM Wiki pattern. Use when user says "知识库", "research wiki", "add paper", "wiki query", "查知识库", or wants to build/query a persistent field map.

Use when experiments complete to judge what claims the results support, what they don't, and what evidence is still missing. Codex MCP evaluates results against intended claims and routes to next action (pivot, supplement, or confirm). Use after experiments finish — before writing the paper or running ablations.

Prepare, validate, and gate AutoDL/HPC research experiments with GitHub deploy-key bootstrap, offline data policy, preflight and smoke gates, FileZilla/SFTP result transfer, and formal-run approval boundaries. Defaults to command preparation over autonomous SSH execution. Use when Codex needs AutoDL, /root/autodl-tmp, remote GPU/HPC smoke tests, preflight_autodl.py, run_autodl_smoke.sh, deploy keys, formal suite gating, or safe download/audit workflows.

Opt-in SSH job queue for multi-seed/multi-config ML experiments with OOM-aware retry, stale-screen cleanup, and wave-transition race prevention. In the lightweight AutoDL-first pack, use only when the user explicitly asks for SSH queue orchestration or approves a prepared command block.

Prepare and gate ML experiment execution with AutoDL/HPC preferred for heavy compute. Runs only local validation/tiny smoke checks by default, prints SSH/remote command blocks for approval, and keeps Vast.ai/Modal as opt-in alternatives. Use when user says "run experiment", "deploy to server", "跑实验", or needs to launch training jobs.

Persistent research knowledge base that accumulates papers, ideas, experiments, claims, and their relationships across the entire research lifecycle. Inspired by Karpathy's LLM Wiki pattern. Use when user says "知识库", "research wiki", "add paper", "wiki query", "查知识库", or wants to build/query a persistent field map.

Use when experiments complete to judge what claims the results support, what they don't, and what evidence is still missing. A secondary Codex agent evaluates results against intended claims and routes to next action (pivot, supplement, or confirm). Use after experiments finish — before writing the paper or running ablations.

Lightweight experiment bridge between idea planning and review. Reads EXPERIMENT_PLAN.md, implements experiment code, runs local validation only, prepares AutoDL/HPC gated execution, and writes prompt-only code-review handoffs. Use when user says "实现实验", "implement experiments", "bridge", "从计划到跑实验", "deploy the plan", or has an experiment plan ready to execute.

Prepare and supervise a clean formal experiment rerun on AutoDL/HPC after smoke tests, with conservative resource planning and safe process/session hygiene. Use when the user says "formal rerun", "AutoDL final run", "HPC experiment gate", "screen重跑", "clean final run", or "正式实验重跑".

Search and analyze research papers, find related work, summarize key ideas. Use when user says "find papers", "related work", "literature review", "what does this paper say", or needs to understand academic papers.

Build and maintain a clean separation between public artifact packages and internal writing/advisor packages. Use when the user says "artifact package", "paper package", "writing package", "public artifact vs internal package", "生成投稿包", or "生成写作用包".

Package a local LaTeX paper tree into an Overleaf-uploadable zip without using the Overleaf Git bridge. Use when the user wants to move a local paper to Overleaf by manual upload, needs an uploadable archive for collaborators, or wants a pre-upload audit of missing figures, bib files, and local-only build junk.

Compile LaTeX paper to PDF, fix errors, and verify output. Use when user says "编译论文", "compile paper", "build PDF", "生成PDF", or wants to compile LaTeX into a submission-ready PDF.

Audit paper/artifact figures and tables after final results are generated, focusing on visual clarity, caption consistency, numeric provenance, and submission-readiness. Use when the user says "check figures", "visualization audit", "图挤了", "caption/table audit", "artifact figure QA", or "投稿前检查图表".