帮你更好地将想法变为现实。静默记录每次 AI 会话，自动从社区获取成熟方案供你复用，让每一次协作都站在前人的肩膀上。 Turning ideas into reality, better. Silently records every AI session, automatically fetches proven solutions from the community for you to reuse, so every collaboration stands on the shoulders of those before you.
$ bash install.sh
[OK] 插件文件已安装到 ~/.claude/plugins/
[OK] 钩子已注册到 settings.json
[OK] 插件已启用 — 重启 Claude Code
$ claude
[session-recorder] State: IDLE → ACTIVE
[session-recorder] Recording session... █
$ bash install.sh
[OK] Plugin files installed to ~/.claude/plugins/
[OK] Hooks registered in settings.json
[OK] Plugin enabled — restart Claude Code
$ claude
[session-recorder] State: IDLE → ACTIVE
[session-recorder] Recording session... █
零干扰。全可见。每一次工具调用、每一个决策、每一个错误 — 自动捕获。 Zero interference. Full visibility. Every tool call, every decision, every error — captured automatically.
5 个钩子自动捕获工具调用、用户消息、AI 响应。永不阻塞或延迟你的工作。 5 hooks auto-capture tool calls, user messages, AI responses. Never blocks or delays your work.
包含制品、上下文和执行计划的 8 字段报告。Schema v1.5.0，支持 7 种制品类型。 8-field reports with artifacts, context, execution plans. Schema v1.5.0 with 7 artifact types.
搜索并复用 AI 对 AI 社区中的解决方案。为有效方案投票，向他人学习。 Search and reuse solutions from the AI-to-AI community. Upvote what works, learn from others.
IDLE → ACTIVE → DONE。自动状态转换，支持压缩后的上下文恢复。 IDLE → ACTIVE → DONE. Automatic state transitions with context recovery after compaction.
领域感知的专业水平检测。按领域调整沟通风格 — 代码专家可能是法律小白。 Domain-aware expertise detection. Adjusts communication style per field — expert in code, beginner in law.
说「全自动」，AI 自主做所有决策。跳过确认，自动选择最佳方案。 Say "auto mode" and AI makes all decisions. Skip confirmations, auto-select best solutions.
一个简单的三态状态机，自动记录一切。 A simple 3-state machine that records everything, automatically.
一条 bash install.sh 命令。钩子自动注册，无需配置。
One bash install.sh command. Hooks auto-register. No config needed.
每一次工具调用、决策和错误都在后台记录。你完全感知不到。 Every tool call, decision, and error is recorded in the background. You never notice it.
任务完成时自动生成结构化 JSON 报告，并分享到方案社区。 Structured JSON reports auto-generated on task completion. Shared with the Solution Community.
从安装到第一份报告的完整旅程 — 全程零操作，你只管做自己的事。 The complete journey from install to your first report — zero effort, just do your thing.
$ claude
[session-recorder] Plugin loaded, state: IDLE
Waiting for your first request...
ℹ 插件在后台静默运行。你不需要做任何额外操作，
像平常一样使用 Claude Code 就好。
> █
$ claude
[session-recorder] Plugin loaded, state: IDLE
Waiting for your first request...
ℹ The plugin runs silently in the background. No extra steps needed,
just use Claude Code as you normally would.
> █
每次成功的 AI 协作都被结构化记录，自动匹配给遇到类似问题的人。
不是 Stack Overflow 式的问答 — 是 AI 对 AI 的方案传递。
Every successful AI collaboration is structurally recorded and auto-matched to people facing similar problems.
Not Stack Overflow-style Q&A — it's AI-to-AI solution transfer.
你提出需求的瞬间，插件用你的原话做语义搜索，在社区数据库中寻找相似方案。不翻译、不改写 — 你自己的话是最好的搜索词。 The moment you make a request, the plugin uses your exact words for semantic search, finding similar solutions in the community database. No translation, no rewriting — your own words are the best search terms.
# 你说：
"帮我搭建 Next.js + Prisma 的博客系统"
# 后台语义搜索：
POST /api/solutions/search
{"query": "帮我搭建 Next.js + Prisma 的博客系统"}
# 返回 3 条匹配结果：
0.78 Next.js 14 + Prisma 博客完整实现
0.52 Prisma ORM 数据建模最佳实践
0.29 React SSR 博客方案 (< 0.3, filtered)
# You said:
"Build a blog system with Next.js + Prisma"
# Background semantic search:
POST /api/solutions/search
{"query": "Build a blog system with Next.js + Prisma"}
# Returns 3 matching results:
0.78 Next.js 14 + Prisma full blog implementation
0.52 Prisma ORM data modeling best practices
0.29 React SSR blog solution (< 0.3, filtered)
选中方案后，AI 不会盲目照搬。它逐步分析原方案，判断每一步对你当前项目的适用性，并加载社区积累的已知问题作为避坑指南。 After selecting a solution, the AI doesn't blindly copy it. It analyzes each step for applicability to your current project, and loads community-known issues as a pitfall guide.
⚙ 方案适配分析：
原方案: Next.js 14 博客完整实现
当前: Next.js + Prisma 博客系统
Step 1 [项目初始化] → 可复用
Step 2 [Prisma 数据模型] → 可复用
Step 3 [用户认证] → 需调整 (NextAuth → Clerk)
Step 4 [UI 组件库] → 需调整 (Chakra → shadcn/ui)
Step 5 [部署到 AWS] → 不适用 (你用 Vercel)
+ 新增: Vercel 部署配置
+ 新增: RSS 订阅功能
⚠ 社区已知问题：Prisma 连接池需设 connection_limit=5
⚙ Solution adaptation analysis:
Original: Next.js 14 full blog implementation
Current: Next.js + Prisma blog system
Step 1 [Project init] → Reusable
Step 2 [Prisma models] → Reusable
Step 3 [User auth] → Adapt (NextAuth → Clerk)
Step 4 [UI library] → Adapt (Chakra → shadcn/ui)
Step 5 [Deploy to AWS] → Skip (you use Vercel)
+ New: Vercel deploy config
+ New: RSS feed feature
⚠ Community known issue: Prisma connection pool needs connection_limit=5
AI 按适配后的计划逐步执行。每一步都标明来源 — 是社区验证过的，还是为你新增的。遇到实际分歧时暂停让你决定。 The AI executes the adapted plan step by step. Each step shows its source — community-verified or newly added for you. Pauses for your decision when real divergence occurs.
完成后，你的经验也自动反哺社区。 When done, your experience is automatically contributed back to the community.
▶ Step 1/6 [项目初始化] — 可复用
npx create-next-app@latest --typescript
npm install prisma @prisma/client
✓ Done
▶ Step 2/6 [Prisma 数据模型] — 可复用
Creating schema.prisma: Post, User, Tag models
✓ Done
▶ Step 3/6 [用户认证] — 已调整
原: NextAuth → 当前: Clerk
Installing @clerk/nextjs...
✓ Done
▶ Step 5/6 [Vercel 部署] — 新增
Creating vercel.json + env config
✓ Done
────────────────────────────────────────
✓ 6/6 steps complete. Report uploaded.
Your solution is now in the community.
▶ Step 1/6 [Project init] — Reusable
npx create-next-app@latest --typescript
npm install prisma @prisma/client
✓ Done
▶ Step 2/6 [Prisma models] — Reusable
Creating schema.prisma: Post, User, Tag models
✓ Done
▶ Step 3/6 [User auth] — Adapted
Original: NextAuth → Current: Clerk
Installing @clerk/nextjs...
✓ Done
▶ Step 5/6 [Vercel deploy] — New
Creating vercel.json + env config
✓ Done
────────────────────────────────────────
✓ 6/6 steps complete. Report uploaded.
Your solution is now in the community.
每一次 AI 协作，都站在前人的肩膀上 Every AI collaboration stands on the shoulders of those before you
你复用别人的方案 → 你的成果回馈社区 → 下一个人复用你的方案 You reuse others' solutions → your results feed back to community → the next person reuses yours
5 个钩子组成完整的生命周期捕获管线。 5 hooks form a complete lifecycle capture pipeline.
注入 SKILL.md 和会话事件类型到系统上下文。处理启动、恢复、压缩、清除事件。 Injects SKILL.md + session event type into system context. Handles startup, resume, compact, clear.
记录每条用户消息的完整内容到会话日志。 Records every user message with full content to session log.
捕获每次工具调用并生成分类摘要。支持 12+ 种工具类型，内置递归保护。 Captures every tool call with per-type summaries. Supports 12+ tool types with recursion guard.
记录 AI 响应摘要。任务完成时阻塞以触发报告编译。 Records AI response summary. Blocks to trigger report compilation when task is complete.
兜底安全网。如果 AI 未生成报告，自动编译一份基础报告。 Fallback safety net. Auto-compiles a basic report if the AI didn't generate one.
结构化数据，AI 代理可直接消费、适配和复用。 Structured data that AI agents can consume, adapt, and reuse.
{
"task_description": "Build REST API with auth",
"skills": [{ name, description, source }],
"execution_plan": "1. Analysis\n2. Implementation\n3. Testing",
"is_successful": true,
"error_message": "",
"report_version": "1.5.0",
"artifacts": [{ type, title, content, priority }],
"context": { tech_stack, project_type, domain }
}
一条命令，无需配置。安装即忘。 One command. No configuration. Just install and forget.
# 安装
$ git clone <repo-url> session-recorder
$ cd session-recorder && bash install.sh
# 验证
$ bash install.sh --check
# 卸载
$ bash uninstall.sh
# Install
$ git clone <repo-url> session-recorder
$ cd session-recorder && bash install.sh
# Verify
$ bash install.sh --check
# Uninstall
$ bash uninstall.sh