Reviews a popular-science article for academic and conceptual correctness. Checks whether explanations of ML/AI, programming, or other technical concepts align with academic consensus and current research literature. Verifies that analogies don't create fundamental misconceptions, citations represent the literature accurately, and framings aren't oversimplified to the point of being wrong. Returns a structured Markdown report with Critical/Warning/Minor issues. Does NOT verify raw facts (dates, numbers, etc.) — that's the fact-checker's job.
Reviews a popular-science article for factual accuracy. Verifies specific factual claims — dates, names, numbers, version numbers, API signatures, product capabilities, quoted statements — against authoritative sources via WebSearch and WebFetch. Returns a structured Markdown report with Critical/Warning/Minor issues. Does NOT judge writing style, tone, or academic framing.
Run academic-reviewer independently on a single article — audits concept explanations, analogies, and academic framing for correctness against research/literature consensus (does NOT verify raw facts — that's fact-checker's job). Auto-bootstraps .composir/<slug>-plan.md if missing by reading the article and proposing an authoritative-source whitelist for user confirmation. Use when you want to audit academic rigor without running the full review-cycle (no fact-checker in the loop). Iterates up to 5 rounds fixing Critical issues.
Drive an interactive brainstorming session for popular science writing. Use when starting a new article or series, discussing topic, angle, target audience, whether to make a series, series structure, article count, and logical flow. Produces a brainstorm.md document summarizing all decisions.
Mechanically validate the format spec of a Chinese or English article — title length, subtitle/SEO description length, SEO tag count, series prefix, first sentence format, 摘要/Summary format, and other rules from the writing-style skill. Use after writing or translating an article to catch format violations before publishing. Reports pass/fail with specific violations.
Run fact-checker independently on a single article — verifies factual claims (dates, names, versions, API signatures, quotes) against authoritative sources. Auto-bootstraps .composir/<slug>-plan.md if missing by reading the article and proposing an authoritative-source whitelist for user confirmation. Use when you want to verify facts without running the full review-cycle (no academic-reviewer in the loop). Iterates up to 5 rounds fixing Critical issues.
Generate a structured plan.md from an existing brainstorm.md for popular science writing. Use after /composir:brainstorm completes, or when the user has a brainstorm document and wants to produce an executable writing plan with article details, research needs, and progress tracking.
Uses power tools
Uses Bash, Write, or Edit tools
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.
Claude Code plugin for the popular-science writing workflow: brainstorm → plan → write → review.
本插件通过 AgenticFish marketplace 分发。
# 在 Claude Code 里
/plugin marketplace add AgenticFish/marketplace
/plugin install composir@agenticfish
/brainstorm驱动一次交互式头脑风暴,和你讨论:写什么、单篇 or 系列、系列结构、关键概念、类比等。还会和你一起列出本主题的权威源白名单(核查阶段用),并让你确认一个 slug(例:git-worktree、ai-journey-android-skills),产出 .composir/<slug>-brainstorm.md。如果主题涉及分析某代码库,会要求你把仓库 clone 到本地并记录 commit/tag。
/plan读取 .composir/<slug>-brainstorm.md,产出结构化的 .composir/<slug>-plan.md——包含每篇文章的详细规划、关键术语、核查要点、权威源白名单、代码库位置(如适用)、进度追踪表。slug 和 brainstorm 一致。
writing-style自动加载的风格指南。user-invocable: false——Claude 在涉及写作任务时自动加载,不需要你手动调用。
/review-cycle [article-path]对写好的草稿执行完整核查循环:fact-checker 和 academic-reviewer 两个 agent 并行审查,发现 Critical 自动修订,再次核查——最多 5 轮。
几个关键设计:
每轮的报告和快照都保存在 .composir/ 下,文件名带 <article-slug>-review-*-iter<N>.md 和 <article-slug>-snapshot-iter<N>.md 前缀。
fact-checker独立事实核查员。验证具体事实——日期、名字、版本号、API 签名、引用——用 WebSearch 和 WebFetch 查一手源。只看事实,不管风格和学术框架。
academic-reviewer学术/概念审查员。检查对概念和原理的解释是否符合学术共识,类比是否误导,有无过度简化到错的地步。不管事实细节。
/fact-check [article-path](0.5.1+)对单篇文章只跑 fact-checker,不走 full review-cycle。文章没有 .composir/<slug>-plan.md 时自动 bootstrap——读文章、推断权威源白名单候选、你确认后写最小 plan.md。迭代上限 5 轮,规则同 review-cycle。适合:只想验事实不要学术审、快速核刚改的段落、对没走过 brainstorm/plan 的老文章做事后体检。
/academic-check [article-path](0.5.1+)对单篇文章只跑 academic-reviewer,不走 full review-cycle。同样支持 plan.md 缺失时 bootstrap。适合:审概念框架和类比是否误导、担心过度简化到"读者会被引到错误方向"。迭代规则和 fact-check 对称。
/research [query]写作时的精确术语查询。结构化 WebSearch + WebFetch,过滤权威源,返回简洁摘要附上 URL。比直接 WebSearch 可靠,适合写作中遇到不确定的点。
/check-format [article-path]机械的格式合规检查——字符数(摘要 100-120、英文 Title < 100、Subtitle < 140、SEO Description < 150)、Tags 数量、系列前缀、系列声明等。用 Python 精确计数,不目测。
/translate-to-english [Chinese-article-path]中译英。不是逐字翻译——保持结构、类比、术语,用自然英语表达。自动生成 Title/Subtitle/SEO Description/Tags/Summary,完成后自动调 check-format 验证,不通过自动重试最多 3 次。不自动触发,仅在用户明确说"写英文"时运行。
/brainstorm ← 主题、读者、形式、结构、slug
↓ .composir/<slug>-brainstorm.md
/plan ← 结构化 plan
↓ .composir/<slug>-plan.md(用户审核)
[写作,中途可随时 /research <term> 查资料]
↓ 中文草稿(放在系列根目录)
/review-cycle article.md ← 自动 fact + academic 核查 + 修订循环
↓ 通过后 .composir/<slug>-plan.md 标为"定稿"
/check-format article.md ← 发布前机械格式检查
↓ 用户说"写英文"
/translate-to-english article.md
↓ 英文草稿(含自动 check-format)
/review-cycle english.md ← 可选:英文也跑一遍核查
所有 .composir/ 下的文件都带slug 前缀,这样同一个目录下可以并存多个独立的 brainstorm/plan(适合把多篇单篇文章放一起)。
系列的典型布局(slug = 系列 slug):
<系列目录>/
├── 01-xxx.md ← 文章本体
├── 02-yyy.md
├── 01-xxx-english.md
└── .composir/
├── ai-journey-android-skills-brainstorm.md
├── ai-journey-android-skills-plan.md
├── 01-xxx-snapshot-iter1.md ← 每轮 agent 看到的版本
├── 01-xxx-snapshot-iter2.md
├── 01-xxx-review-fact-iter1.md
├── 01-xxx-review-academic-iter1.md
├── 01-xxx-review-fact-iter2.md
└── 01-xxx-review-academic-iter2.md
多个单篇共享目录(每个单篇自己的 slug):
<随笔目录>/
├── git-worktree.md ← 文章本体
├── ssh-config-tricks.md
└── .composir/
├── git-worktree-brainstorm.md
├── git-worktree-plan.md
├── git-worktree-review-fact-iter1.md
├── ssh-config-tricks-brainstorm.md
└── ssh-config-tricks-plan.md
所有外部 URL 抓取走 .composir/.cache/ 三层缓存:
bin/composir-fetch <url> 用 curl 存原始 HTML,只缓存 2xx 响应;后续 Read 复用<hash>.wf.md,下一轮 agent 可命中已有 Q/A/composir:review-cycle 多轮迭代、或 brainstorm → fact-check 跨 skill 的同 URL 复用.composir/.cache/),系列之间互不共享# 某个 URL 的内容过时了,想重抓
rm .composir/.cache/<hash>-<hint>.html
# 或清空整个系列的缓存
rm -rf .composir/.cache/
composir-fetch: permission denied → chmod +x ${CLAUDE_PLUGIN_ROOT}/bin/composir-fetchcomposir-fetch: need shasum or sha1sum → 系统缺 hash 工具;macOS/Linux 默认都有,极简容器里装 coreutils.composir/.gitignore 应含 .cache/,没有的话 composir-fetch 会自动补npx claudepluginhub agenticfish/marketplace --plugin composirComplete creative writing suite with 10 specialized agents covering the full writing process: research gathering, character development, story architecture, world-building, dialogue coaching, editing/review, outlining, content strategy, believability auditing, and prose style/voice analysis. Includes genre-specific guides, templates, and quality checklists.
Upstash Context7 MCP server for up-to-date documentation lookup. Pull version-specific documentation and code examples directly from source repositories into your LLM context.
Comprehensive startup business analysis with market sizing (TAM/SAM/SOM), financial modeling, team planning, and strategic research
v9.44.1 — Patch release for Gemini environment/version detection and qwen auth gating. Run /octo:setup.
Comprehensive .NET development skills for modern C#, ASP.NET, MAUI, Blazor, Aspire, EF Core, Native AOT, testing, security, performance optimization, CI/CD, and cloud-native applications
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.