By skymanbp
Scientific paper writing + review + ideation pipeline for top-tier journals (ApJ/MNRAS/PRD/JCAP-class). Ships 8 skills (paper, paper-review, figure-review, paper-style, brainstorm, mainline, paper-attack-tree, final-review) and 7 corpus-driven style tools. See CHANGELOG.md for per-version history.
全自动辐射状探索器。可用于发散式寻找研究 ideas,也可用于穷尽地寻找解决某个切实难题的途径。以"系统发生树(phylogenetic tree)"为数据模型:root = 起点(topic / 问题 / 当前研究状态);node = 一个想法;depth = 发散层数;width = 最终结果数。每一节点用多种视角(first-principles / 反演 / 跨学科迁移 / 对抗 / 约束变换 / 尺度外推 / office-hours / contrarian / 失效驱动 / high-risk / 元层 等)穷尽 brainstorm,每个分支完整严谨推导(数学/物理/逻辑/文献核对/可行性/可证伪性),递归发散直至每一最深叶节点完整推进。启动时强制 §2.0 glossary grill 预热(与 mattpocock-skills:grill-with-docs 同源),把 root 节点术语锁到项目 FACTS.md。默认 width / depth / rounds **全部不限**,由收敛判据终止。**严禁** "defer / 因成本限制 / future work / TODO" 等推脱式不完整结果。强制 cc-enslaver 七规则全程证据可追溯。Use when 用户说 "brainstorm" / "发散思考" / "找研究方向" / "怎么解决这个难题" / "explore options" / 想穷尽某个研究问题的解法 / 论文 motivation 阶段需要 radial 探索。
Review paper figures at print-realistic DPI for readability, consistency, and journal standards. Renders every page of the compiled PDF at 150 DPI (simulating a reader viewing at 100% zoom on a standard monitor) and checks each figure systematically. Use when 用户说 "查图" / "figure-review" / "图能不能读" / "图够不够清楚" / "导出的 PDF 图字号太小" / 投稿前图表 audit / 用 `\includegraphics` 的 paper.tex 编完后 final 检查。
最终审阅编排器。完整使用 5 个 skill 作为框架——paper(写作标准基线,主代理本进程加载)/ paper-review(A–R 全维度 checklist 审查,传 `--no-isolated-mpr` 由父级接管 MPR)/ figure-review(图表 150-DPI 关)/ mainline(结构 spine)/ paper-attack-tree(adversarial radial critique),并在 §3.6 由 final-review 主代理在隔离 worktree 中直接调一次 user-level modern-physics-review 作为独立 5th 子代理(修复嵌套 sub-agent bug)。**每个子代理都在 isolation=worktree 中独立运行**,cold-read 论文消除框架偏见。每轮 merge 所有 issue 后修复,循环直至**连续 2 轮 5 个 isolated skill (paper-review/figure/mainline/attack/mpr) 均 0 issue**(稳态收敛)。**严禁** "基本干净 / 大致收敛 / 剩下都是 minor / 用户没时间提前结束"。ITER_BUDGET 默认 10 轮,触顶 BREAK_WITH_USER_DECISION(不允许偷偷宣布完成)。Use when 用户说 "final-review" / "投稿前最终审" / "总审" / "运行所有审查 skill" / submission readiness check。
主线增强 / 叙事脊柱锐化器。完整阅读全文(禁止 grep-only / 记忆 / 猜测),从 7 个正向维度(主线锐化 / 语言精简 / 叙事结构 / 隔离可读性 / 推导完整 / 逻辑合理 / 主线串联)和 8 个反向维度(定义模糊 / 主线分散 / 多而不精 / 章节无关联 / 结构不清 / 缺学术叙事 / 上下文不统一 / 低信息量形容语句)双向审查并自动修复;专门处理 brainstorm 发散后的观点收敛与叙事整合;强制隔离上下文 cold-read 可读性二审;零问题收敛硬闭环。与 paper-review 互补(paper-review 管 per-claim 正确性,本 skill 管结构层 spine)。Use when 用户说 "主线不清" / "叙事乱" / "结构层审查" / "mainline" / "整合 brainstorm 输出" / "章节衔接有问题" / "story 不顺" / 想做 spine-level 审查(per-claim 已经 paper-review 过)。
用 brainstorm 的辐射状探索方法对论文做 critique tree 审查。每个 node = 一个 critique;用 12 条 framing pass(first-principles / 反演 / 跨学科 reviewer / 对手红队 / 约束变换 / 尺度外推 / 替换 / office-hours / contrarian / 失效驱动 / high-risk 致命攻击 / 元层)从多角度攻击每条 claim;每个 critique 必须完整推进至 CONFIRMED(有 file:line 证据 + 具体修复方案)/ REFUTED(有 file:line 证据证明论文已处理)/ MARGINAL(依赖解读,列入作者判断)。**严禁** "defer / NEEDS-MORE-INFO 滞留 / 因成本 / 因时间 / future work / 可能存在该问题待确认" 等推脱式不完整 verdict;递归发散直至无新颖增益;强制 cc-enslaver 七规则全程证据可追溯。与 paper-review 互补(paper-review 是预设 checklist 静态审查;本 skill 是开放式 adversarial radial 探索)。Use when 用户说 "attack tree" / "adversarial review" / "对抗审查" / "找 reviewer 会挑什么刺" / "audit this claim" / 用 paper-review 跑完仍想 open-ended 攻击关键 claim / rebuttal 准备阶段。
Own this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimOwn this plugin?
Verify ownership to unlock analytics, metadata editing, and a verified badge. GitHub access is read-only (username + org membership).
Sign in to claimBased on adoption, maintenance, documentation, and repository signals. Not a security audit or endorsement.
A Claude Code plugin for writing and reviewing scientific papers at the ApJ / MNRAS / PRD / JCAP / Nature-Physics level.
| Skill | Purpose |
|---|---|
paper | Writing standards (formula conventions, citation rules, anti-AI-ism hard rules, forward-narrative structural-update rule, key references). Invoke before drafting. |
paper-review | Forced source-traceable review across A–R dimensions (math / physics / logic / language / structure / citations / data / interfaces / redundancy / reproducibility / modern-physics-review / systemic inconsistency / adversarial 3-pass / drift sweep / process-artifact removal / draft language / citation precision / glossary alignment). Zero-issue convergence hard loop; isolated-context MPR final verification. |
figure-review | Renders compiled PDF at 150 DPI and reviews each figure as a human reader would actually see it (axis font sizes, legend readability, float placement). |
paper-style | Loads a corpus-distilled style profile (top-journal + mentor + high-value-reference papers) plus retrieves section-typed exemplar paragraphs. Constrains writing/rewriting to match observed human-academic patterns. |
brainstorm | Fully-automated radial research-direction explorer (phylogenetic-tree model). 12 framings (first-principles / inversion / cross-disciplinary / adversarial / constraint-shift / scale extrapolation / substitution / office-hours / contrarian / failure-driven / high-risk-high-reward / meta-self-audit). Recurses on promising leaves until convergence. §2.0 glossary grill prelude locks root-node terms to FACTS.md. |
mainline | Structural narrative-spine reinforcer. Full-read audit on 7 positive + 8 negative structural dimensions; mandatory isolated-context cold-read 7-question readability sub-agent. Complements paper-review (per-claim correctness) by covering spine-level issues. |
paper-attack-tree | brainstorm's radial methodology applied to critique. Each node = one critique attacked by 12 framing passes; every leaf resolved to CONFIRMED / REFUTED / MARGINAL with file:line evidence. No NEEDS-MORE-INFO defer. Complements paper-review (static checklist) with open-ended adversarial coverage. |
final-review | 5-skill orchestrator for pre-submission final pass. Runs paper-review / figure-review / mainline / paper-attack-tree / modern-physics-review each in its own isolation: worktree sub-agent every round; loops until consecutive N rounds (default 2) show 0 issues across all 5. ITER_BUDGET 10 rounds. |
See CHANGELOG.md for the per-version evolution.
| Tool | Purpose |
|---|---|
tools/build_profile.py | One-shot orchestrator: extract → train classifier → warm exemplar cache. |
tools/extract_style.py | Corpus → lexicon / sentence-stats / transition-inventory JSON + style_dossier.md + section-typed exemplar bank. .tex and .pdf (pymupdf blocks). |
tools/retrieve_exemplars.py | Section + topic + field → top-K exemplar paragraphs (sentence-transformers cosine, .npy cache, keyword-overlap fallback). |
tools/ai_ism_lint.py | Tier-graded regex (em-dash, Tier A zero-tolerance, Tier B frequency-capped) + corpus-derived blacklist + opt-in ML classifier (--ai-classifier) + --summary. |
tools/train_ai_ism_classifier.py | Trains a paragraph-level logistic-regression classifier on word 1–2 gram TF-IDF. Positives = corpus, negatives = ai_ism_negatives_handcrafted.txt (extend with extract_md_negatives.py). CV F1 ≈ 0.88 on wgl. |
tools/extract_md_negatives.py | Harvests LLM-drafted prose from your project tree as extra negatives for the classifier. |
tools/ai_ism_negatives_handcrafted.txt | Seed negatives shipped with the plugin (extend by hand or via extract_md_negatives.py). |
See tools/README.md for per-tool roadmap and graceful-degradation behaviour.
# 1. Install Python deps (numpy + pymupdf + sentence-transformers + sklearn).
pip install -r requirements.txt
# 2. Drop corpus papers into the field tier dirs:
# style-corpus/wgl/tier-1-top/ ← top-journal .tex / .pdf
# style-corpus/wgl/tier-2-mentor/ ← mentor's papers
# style-corpus/wgl/tier-3-reference/ ← other high-value references
# 3. Build the profile end-to-end (extract → train → warm cache).
# First run downloads the sentence-transformers model (~80 MB; ~30 s).
python tools/build_profile.py
# 4. Register the plugin in your Claude Code session
claude --plugin-dir <path-to-this-repo>
# 5. Use it
# - skill (Claude does the work) : /sci-paper:paper-style discussion
# - manual exemplar retrieval : python tools/retrieve_exemplars.py --section method --topic "..."
# - lint a draft (regex + ML classifier): python tools/ai_ism_lint.py draft.tex --ai-classifier --summary
paper-style instead of "fine-tuning a model"npx claudepluginhub skymanbp/sci-paper --plugin sci-paperA Claude Code plugin and LLM-agnostic rule pack that enforces systematic thinking, verified citations, and root-cause analysis — eliminating reactive, lazy AI behavior. v0.20 collapses the agent's reply skeleton into a fixed YAML schema (cc-enslaver: block whose field names ARE the existing Stop-hook detection markers, so no detector changed) and adds Stop layer (h): a hard-enforced one-sentence plain-language TL;DR (大白话总结) required on every done-claim reply, plus a 大白话 line on every block reason. v0.19 adds cwd-fallback for project edicts path resolution when CLAUDE_PROJECT_DIR fails to propagate.
Universal radial-tree exploration engine for Claude Code. One `tree` skill + swappable presets (brainstorm / attack / design / code-audit) for divergent ideation, adversarial critique, and design-space exploration. 12 framings × hard-ban-on-incomplete-leaves × stable convergence. See CHANGELOG.md for per-version history.
Persistent memory for Claude Code — automatic save/restore across compactions and sessions. Anti-patch reconcile-on-write, forced PROGRESS.md handoff, live PLAN.md anchor with plan-refiner / plan-guardian subagents, FTS5 search, AI-judged extraction with Haiku + local Ollama fallback.
A growing collection of Claude-compatible academic workflow bundles. Covers scientific figures, manuscript writing and polishing, reviewer assessment, citation retrieval, data availability, paper reading, literature search, response letters, paper-to-PPTX conversion, and evidence-grounded Chinese invention patent drafting. Rules are organized as reusable skill folders with explicit workflows and quality checks.
Core skills library for Claude Code: TDD, debugging, collaboration patterns, and proven techniques
Harness-native ECC plugin for engineering teams - 67 agents, 271 skills, 92 legacy command shims, reusable hooks, rules, MCP conventions, and operator workflows for Claude Code plus adjacent agent harnesses
UI/UX design intelligence. 67 styles, 161 palettes, 57 font pairings, 25 charts, 15 stacks (React, Next.js, Vue, Svelte, Astro, SwiftUI, React Native, Flutter, Tailwind, shadcn/ui, Nuxt, Jetpack Compose). Actions: plan, build, create, design, implement, review, fix, improve, optimize, enhance, refactor, check UI/UX code. Projects: website, landing page, dashboard, admin panel, e-commerce, SaaS, portfolio, blog, mobile app. Elements: button, modal, navbar, sidebar, card, table, form, chart. Styles: glassmorphism, claymorphism, minimalism, brutalism, neumorphism, bento grid, dark mode, responsive, skeuomorphism, flat design. Topics: color palette, accessibility, animation, layout, typography, font pairing, spacing, hover, shadow, gradient.
Comprehensive skill pack with 66 specialized skills for full-stack developers: 12 language experts (Python, TypeScript, Go, Rust, C++, Swift, Kotlin, C#, PHP, Java, SQL, JavaScript), 10 backend frameworks, 6 frontend/mobile, plus infrastructure, DevOps, security, and testing. Features progressive disclosure architecture for 50% faster loading.
Tools to maintain and improve CLAUDE.md files - audit quality, capture session learnings, and keep project memory current.