[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"skill:01M76W5CWR8MABE0XPZVRXFMDK":3,"skill-same:01M76W5CWR8MABE0XPZVRXFMDK":46},{"skillNo":4,"name":5,"nameZh":6,"description":7,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":16,"categories":17,"keywords":18,"publishedAt":19,"linkOnly":20,"markdownContent":21,"markdownBody":22,"frontMatter":23,"refName":24,"commitSha":25,"skillDirectory":26,"forksCount":14,"repositoryHtmlUrl":27,"repositoryHomepageUrl":6,"ownerHtmlUrl":28,"ownerBlogUrl":29,"ownerDescription":30,"updatedAt":31,"install":32,"resources":45},"01M76W5CWR8MABE0XPZVRXFMDK","xray",null,"Understand what just happened under the hood - learn AI by seeing it in action","hemory\u002Famp","hemory","https:\u002F\u002Favatars.githubusercontent.com\u002Fu\u002F21179881?v=4","01M6D31HFJ391DN8N0JP8AQGFE","Hemory Phifer","01M758Z6KK3NGD8654Y5CT668C",0,"MIT","https:\u002F\u002Fgithub.com\u002Fhemory\u002Famp\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fxray\u002FSKILL.md",[],[],"2026-11-12 13:55:43",false,"---\nname: xray\ndescription: Understand what just happened under the hood - learn AI by seeing it in action\n---\n\n# \u002Fxray - The AI Education Experience\n\n**Default behavior:** Explain what just happened in THIS conversation — the specific tools, files, and context that were used, with educational explanations of WHY.\n\n**With qualifiers:** Deep dive into specific topics (AI fundamentals, Amp architecture, how to extend).\n\n---\n\n## Philosophy\n\nThe best way to learn is by examining what just happened, not abstract concepts. When you run `\u002Fxray`:\n\n1. **Show the work** - What tools ran, what files were read\u002Fwritten\n2. **Explain why** - Why each action was necessary\n3. **Connect to concepts** - Link specific actions to how AI\u002FAmp works\n4. **Empower extension** - \"You could customize this by...\"\n\n**The outcome:** Users learn AI by examining their own conversations, then go deeper if curious.\n\n**Tone:** Like a senior engineer doing a code review of what just happened. \"Here's what I did and why.\"\n\n---\n\n## Step 1: Determine Mode\n\n### Default (no arguments): \"This Conversation\"\n\nWhen user just says `\u002Fxray` with no qualifier:\n→ **Immediately explain what happened in this conversation**\n→ Don't show a menu, just show the work\n\n### With qualifiers: Specific topics\n\n| Command | What it does |\n|---------|--------------|\n| `\u002Fxray` | (default) Explain what just happened in this chat |\n| `\u002Fxray ai` or `\u002Fxray fundamentals` | First principles: context windows, tokens, statelessness |\n| `\u002Fxray amp` or `\u002Fxray architecture` | How Amp works: CLAUDE.md, hooks, MCPs, skills |\n| `\u002Fxray boot` or `\u002Fxray session start` | The session startup sequence |\n| `\u002Fxray today` | ScreenPipe analysis of your day |\n| `\u002Fxray extend` or `\u002Fxray customize` | How to customize and build on Amp yourself |\n\n---\n\n## DEFAULT MODE: This Conversation (Show Your Work)\n\n**This is the primary mode.** When user runs `\u002Fxray`, explain what happened.\n\n### Step 1: Analyze the conversation\n\nLook at the current conversation and identify:\n1. What files were read\n2. What files were written\u002Fmodified\n3. What tools\u002FMCPs were used\n4. What context was loaded at session start (check for `\u003Cprevious_session_memories>` etc.)\n5. Key decision points in the AI's reasoning\n\n### Step 2: Present as educational walkthrough\n\nOutput format:\n\n```markdown\n## 🔬 X-Ray: What Just Happened\n\nLet me explain what happened under the hood in our conversation.\n\n---\n\n### 📖 The Story of This Chat\n\nHere's what I did, step by step:\n\n**1. Session Started**\nBefore you typed anything, the system loaded:\n- `CLAUDE.md` (~X tokens) — My personality, your preferences, available tools\n- Session memories from earlier today — [list what was injected]\n- [Any other context visible from hooks]\n\n**Why this matters:** Without those session memories, I'd have no idea about \n[specific context]. That came from a **session hook** that runs before you type.\n\n**2. You Asked: \"[summarize their first message]\"**\n\nTo respond, I:\n- Read `[file 1]` — Because [reason]\n- Read `[file 2]` — To check [what]\n- Used [tool\u002FMCP] — To [action]\n\n**Why these files?** Your CLAUDE.md tells me where to look. It says [quote relevant \ninstruction]. Without those instructions, I wouldn't know your file structure.\n\n**3. [Continue for each major exchange in the conversation]**\n\n---\n\n### 🔧 Tools Used This Session\n\n| Tool | Times Used | What For |\n|------|------------|----------|\n| `read` | X | Reading vault files |\n| `write` | X | Creating new content |\n| `edit` | X | Updating existing files |\n| `bash` | X | Running commands |\n| [other tools] | X | [purpose] |\n\n**Why tools matter:** Without tools, I can only generate text. Tools let me \nactually DO things — read your files, write new ones, check your calendar, etc.\n\n---\n\n### 💡 Key Insight: The Context Window\n\nEverything I \"know\" right now is what's in the **context window**:\n\n```\n┌─────────────────────────────────────────────────┐\n│  What I can see (the context window)            │\n├─────────────────────────────────────────────────┤\n│ ✓ CLAUDE.md (loaded at start)                   │\n│ ✓ Session memories (injected by hooks)          │\n│ ✓ Our full conversation so far                  │\n│ ✓ Every file I've read this session             │\n│ ✓ Every tool result I've received               │\n├─────────────────────────────────────────────────┤\n│ ✗ Files I haven't read                          │\n│ ✗ Previous sessions (unless summarized)         │\n│ ✗ Anything not loaded into context              │\n└─────────────────────────────────────────────────┘\n```\n\nIf something isn't in this window, I literally don't know it exists.\n\n---\n\n### 📁 Files Touched This Session\n\n**Read (loaded into my context):**\n- `[file]` — [why it was needed]\n- [continue for each file]\n\n**Written\u002FModified:**\n- `[file]` — [what was created\u002Fchanged]\n- [continue for each file]\n\n**Why this matters:** Every file I read uses \"tokens\" (space in the context \nwindow). I try to read only what's relevant, not everything.\n\n---\n\n### 🎓 Concepts You Just Saw in Action\n\n| What Happened | The Underlying Concept |\n|---------------|------------------------|\n| I knew about earlier conversations | **Session hooks** inject context at start |\n| I read CLAUDE.md first | **System prompt** loads before anything |\n| I created files with tools | **Tools** let AI take action, not just talk |\n| I followed your file conventions | **CLAUDE.md instructions** define your structure |\n| I didn't know about [X] | **Context limitation** — only know what's loaded |\n\n---\n\n### 🔮 What You Could Customize\n\nBased on what just happened:\n\n**1. Add to CLAUDE.md**\nIf you want me to always [behavior], add it to your `USER_EXTENSIONS` block.\n\n**2. Create a Skill**\nIf you do [this workflow] often, create a skill to automate it.\n\n**3. Add a Session Hook**\nIf you always want [context] loaded at start, write a hook that injects it.\n\n---\n\n### 🎓 Want to Go Deeper?\n\n- `\u002Fxray ai` — First principles: context windows, tokens, statelessness\n- `\u002Fxray amp` — Full architecture: CLAUDE.md, hooks, MCPs, skills, vault\n- `\u002Fxray boot` — The session startup sequence in detail\n- `\u002Fxray extend` — Step-by-step guide to customizing\n\nOr ask anything: \"How do hooks work?\", \"What's an MCP?\", \"How can I add my own tools?\"\n```\n\n---\n\n## MODE: AI Fundamentals (`\u002Fxray ai`)\n\n**Purpose:** Teach the core concepts that make everything else make sense.\n\n### Output Format\n\n```markdown\n## 🧠 How AI Actually Works\n\nLet me explain the core concepts that underpin everything Amp does.\n\n---\n\n### The Big Truth: AI Has No Memory\n\n**The fundamental problem:** Every time you start a new chat, Claude has amnesia. \nIt doesn't remember yesterday's conversation. It doesn't know your name, your \nprojects, or what you discussed last week.\n\nThis is true for ALL LLM-based AI — ChatGPT, Claude, Gemini, all of them.\n\n**Why?** LLMs are \"stateless.\" They process text in, text out, with no permanent \nstorage between sessions. Each conversation starts from zero.\n\n**So how does Amp remember things?** We engineer around the limitation. \nThat's what Amp is — clever engineering to give a forgetful AI the *appearance* \nof memory, context, and continuity.\n\n---\n\n### Concept 1: The Context Window\n\nWhen you chat with an AI, there's an invisible \"window\" of text the AI can see:\n\n```\n┌─────────────────────────────────────────────────┐\n│              THE CONTEXT WINDOW                 │\n│  (Everything the AI knows about THIS chat)      │\n├─────────────────────────────────────────────────┤\n│ • System prompt (CLAUDE.md)                     │\n│ • Injected context (session memories, etc.)     │\n│ • Our conversation so far                       │\n│ • Any files I've read                           │\n│ • Tool results (MCP responses)                  │\n└─────────────────────────────────────────────────┘\n          ↓\n     AI generates response based on ALL of this\n```\n\n**Key insight:** The AI can only work with what's IN the context window. \nIf information isn't in there, the AI doesn't know it exists.\n\n**Amp's job:** Get the RIGHT information INTO that window at the RIGHT time.\n\n---\n\n### Concept 2: Tokens (The Currency)\n\nContext windows are measured in \"tokens\" (roughly 4 characters = 1 token).\n\n- Claude's window: ~200,000 tokens (~150,000 words)\n- Your CLAUDE.md: ~4,500 tokens\n- A typical person page: ~500 tokens\n\n**Why it matters:** There's a budget. Load too much context and you hit limits \nor slow things down. Load too little and the AI lacks crucial information.\n\n**Amp's approach:** Selective loading. Don't dump everything — load what's \n*relevant* to what you're asking about.\n\n---\n\n### Concept 3: System Prompts (The Personality)\n\nBefore your first message, there's already text in the context window: \nthe **system prompt**. For Amp, that's `CLAUDE.md`.\n\n```\n┌─────────────────────────────────────────────────┐\n│ CLAUDE.md (loaded before you type anything)     │\n├─────────────────────────────────────────────────┤\n│ \"You are Amp, a personal knowledge assistant\"   │\n│ \"User's pillars: [your pillars]\"                │\n│ \"When tasks are mentioned, use Work MCP\"        │\n│ \"Person pages are in 05-Areas\u002FPeople\u002F\"          │\n│ ... (hundreds more lines of context)            │\n└─────────────────────────────────────────────────┘\n```\n\n**The power move:** Everything in CLAUDE.md influences EVERY response. \nIt's like setting the AI's personality and knowledge before you say hello.\n\n---\n\n### Concept 4: Tools (How AI Does Things)\n\nClaude can't *do* things by default — it can only generate text. But with \n**tools**, it can take actions:\n\n- Read files from your computer\n- Write new files\n- Run commands\n- Query databases\n- Call APIs\n\nIn Amp, tools come from **MCP servers** (Model Context Protocol). Each MCP \ngives Claude new capabilities.\n\n**Without tools:** Claude can only talk.\n**With tools:** Claude can take action on your behalf.\n\n---\n\n### Concept 5: The Conversation Loop\n\nHere's what happens every time you send a message:\n\n```\n1. Your message gets added to context window\n                    ↓\n2. Claude sees: system prompt + history + your message\n                    ↓\n3. Claude decides: should I use a tool?\n   - If yes → Tool runs → Result added to context → Back to step 3\n   - If no → Generate response\n                    ↓\n4. Response appears (and gets added to context for next turn)\n                    ↓\n5. Wait for your next message → Repeat from step 1\n```\n\n**Key insight:** Each turn, Claude sees MORE context (the history grows). \nLong conversations = more context = closer to limits.\n\n---\n\n### So What?\n\nUnderstanding these fundamentals helps you:\n\n1. **Know why Amp loads things at session start** — to get context into the window\n2. **Understand why some info is \"missing\"** — it wasn't loaded into context\n3. **See opportunities to extend** — what else could be loaded? what tools could help?\n\n---\n\n**Want to see Amp's specific architecture?** Try `\u002Fxray amp`\n```\n\n---\n\n## MODE: Amp Architecture (`\u002Fxray amp`)\n\n**Purpose:** Explain the specific building blocks that make Amp work.\n\n### Output Format\n\n```markdown\n## 🏗️ How Amp Works: The Architecture\n\nNow that you understand AI fundamentals, here's how Amp engineers around \nthe limitations to create something useful.\n\n---\n\n### The Core Insight\n\n**Problem:** Claude forgets everything between sessions.  \n**Solution:** Store everything in FILES and reload context intelligently.\n\n```\n┌─────────────────────────────────────────────────┐\n│                YOUR VAULT                        │\n│  (Plain Markdown files YOU own)                 │\n├─────────────────────────────────────────────────┤\n│ • CLAUDE.md - Core context & personality        │\n│ • 05-Areas\u002FPeople\u002F - Relationship memory        │\n│ • 03-Tasks\u002FTasks.md - Work in progress          │\n│ • System\u002FSession_Memory\u002F - Past conversations   │\n│ • System\u002FWork_In_Progress.md - Active projects  │\n└─────────────────────────────────────────────────┘\n            ↓ loaded at session start ↓\n         ┌─────────────────────────────┐\n         │      AI Context Window      │\n         └─────────────────────────────┘\n```\n\n**The magic:** Your knowledge lives in files. AI reads files. \nTherefore, AI can \"remember\" by reading what was written before.\n\n---\n\n### Building Block 1: CLAUDE.md (The Brain)\n\nThe system prompt that defines who Amp is and how it behaves.\n\n**Location:** `\u002FCLAUDE.md` (root of your vault)\n\n**Contains:**\n- Your user profile (name, role, company)\n- Your strategic pillars\n- How Amp should behave\n- What skills are available\n- References to other docs\n\n**Why it matters:** This loads FIRST, before anything else. It shapes every \nresponse. Edit this, and you change how Amp thinks.\n\n**You can customize:**\n```markdown\n## USER_EXTENSIONS_START\n\u003C!-- Your personal additions go here -->\n## USER_EXTENSIONS_END\n```\n\n---\n\n### Building Block 2: Session Hooks (The Boot Sequence)\n\nWhen you start a chat, things happen BEFORE you type anything.\n\n**What hooks do:**\n- Inject session memories from recent conversations\n- Surface relevant context (person info, project status)\n- Pre-load data you'll likely need\n\n**Current hooks in Amp:**\n- Session memories injection\n- Person context surfacing\n- Company context surfacing\n\n**The power:** You can ADD hooks. Want weather loaded at session start? \nYour calendar? Write a hook, and it's there every session.\n\n---\n\n### Building Block 3: MCP Servers (The Toolbox)\n\nMCP = Model Context Protocol. It's how Claude gets superpowers.\n\nEach MCP server gives Claude new abilities:\n\n| Server | What It Does |\n|--------|--------------|\n| **Work MCP** | Task management with deduplication and pillar alignment |\n| **Calendar MCP** | Reads your Apple Calendar |\n| **Granola MCP** | Searches meeting transcripts |\n| **Career MCP** | Career evidence tracking and analysis |\n\n**Why MCPs instead of just reading files?**\n- **Speed:** Pre-process data instead of parsing raw files\n- **Logic:** Implement business rules (like \"can't have 4 P0 tasks\")\n- **Integration:** Connect to external systems\n\n---\n\n### Building Block 4: Skills (The Playbook)\n\nSkills are structured instructions for multi-step workflows.\n\n**Location:** `.claude\u002Fskills\u002F[skill-name]\u002FSKILL.md`\n\n**Examples:**\n- `\u002Fdaily-plan` - Check calendar, review tasks, generate plan\n- `\u002Fmeeting-prep` - Load person context, surface recent discussions\n- `\u002Fxray` - This skill you're running right now!\n\n**Why skills?**\n- **Consistency:** Same workflow every time\n- **Complexity:** Handle multi-step processes\n- **Reusability:** Build once, use forever\n\n---\n\n### Building Block 5: The Vault (The Memory)\n\nEverything persists to Markdown files:\n\n```\nYour Vault\u002F\n├── 00-Inbox\u002F           ← Capture zone\n├── 03-Tasks\u002F           ← Active tasks\n├── 04-Projects\u002F        ← Time-bound work\n├── 05-Areas\u002F           ← Ongoing responsibilities\n│   ├── People\u002F         ← Person pages\n│   └── Companies\u002F      ← Company profiles\n├── 06-Resources\u002F       ← Reference material\n└── System\u002F             ← Amp configuration\n```\n\n**Why plain Markdown?**\n- You own it forever (no vendor lock-in)\n- Works anywhere (Obsidian, VS Code, any editor)\n- AI can read and write it\n- Version control with Git if you want\n\n---\n\n### How It All Connects\n\n```\nSession starts\n     ↓\nHooks fire (inject context)\n     ↓\nCLAUDE.md loads (personality + instructions)\n     ↓\nYou type your message\n     ↓\nClaude reads relevant files (on-demand)\n     ↓\nClaude uses MCPs (if needed)\n     ↓\nResponse generated + files updated\n     ↓\nRepeat for each message\n```\n\n---\n\n### Key Insight: It's All Text\n\nThere's no magic database. No proprietary format. No cloud service.\n\n- CLAUDE.md is text\n- Skills are text\n- Your notes are text\n- Session memories are text\n\nClaude reads text, processes it, generates text, writes text.\n\n**Everything you're experiencing is text transformation.**\n\nUnderstanding this is liberating — you can read, edit, and extend anything.\n\n---\n\n**Want to see the session startup in detail?** Try `\u002Fxray boot`\n**Ready to customize?** Try `\u002Fxray extend`\n```\n\n---\n\n## MODE: Session Boot Sequence (`\u002Fxray boot`)\n\n**Purpose:** Detailed walkthrough of what happens when a session starts.\n\n### Output Format\n\n```markdown\n## 🚀 What Happens When You Start a Chat\n\nLet me walk you through exactly what happens from the moment you open \na new chat to when you type your first message.\n\n---\n\n### The Boot Sequence (In Order)\n\n#### Step 1: Environment Detection\n```\nIs this Claude Desktop? Claude Code? Cursor? PI?\n     ↓\nEach environment has slightly different capabilities\n```\n\n#### Step 2: Settings Load\n```\nRead configuration files\n     ↓\nThis defines which hooks to run and which MCPs are available\n```\n\n#### Step 3: Hooks Fire (CRITICAL)\n```\nPre-session hooks run in order:\n     ↓\n┌─────────────────────────────────────────────┐\n│ Session memory injection                    │\n│ → Reads System\u002FSession_Memory\u002F[date].md     │\n│ → Injects summaries of recent conversations │\n│ → AI \"remembers\" what you discussed before  │\n└─────────────────────────────────────────────┘\n     ↓\n┌─────────────────────────────────────────────┐\n│ Context injectors                           │\n│ → Person context ready when you mention     │\n│   names                                     │\n│ → Company context ready when relevant       │\n└─────────────────────────────────────────────┘\n     ↓\n┌─────────────────────────────────────────────┐\n│ (Your custom hooks could go here)           │\n│ → Whatever YOU want loaded at start         │\n└─────────────────────────────────────────────┘\n```\n\n**This is where the magic happens!** Hooks let you inject ANY context \nbefore the conversation starts.\n\n#### Step 4: CLAUDE.md Loads\n```\nThe entire CLAUDE.md file enters the context window:\n     ↓\n• Your profile (name, role, company)\n• Your strategic pillars  \n• Behavior guidelines\n• Reference to skills and tools\n• User extensions (YOUR customizations)\n```\n\n#### Step 5: MCPs Connect\n```\nMCP servers become available:\n     ↓\n• Calendar MCP → Ready to check schedule\n• Work MCP → Ready to manage tasks\n• Granola MCP → Ready to search meetings\n```\n\n#### Step 6: Ready for Input\n```\nNOW you can type your first message.\n\nThe AI already has:\n• Your personality and preferences (CLAUDE.md)\n• Recent conversation summaries (from hooks)\n• Relevant context (from hooks)\n• Access to tools (MCPs)\n```\n\n---\n\n### What This Means for You\n\n**The AI doesn't start from zero.** By the time you type your first message, \nClaude already knows who you are, what you've been working on, and what \ntools are available.\n\n**The boot sequence IS the memory system.** It's how a stateless AI \n\"remembers\" who you are.\n\n---\n\n### 💡 Customization Opportunity\n\n**What else could be loaded at session start?**\n\nIdeas for hooks:\n1. Today's calendar events\n2. Current sprint goals\n3. Recent Git commits\n4. Weather for outdoor planning\n5. Your daily intention\n\n**The pattern:** Anything you want the AI to \"know\" at the start of every \nsession can be injected via a hook.\n\n---\n\n**Ready to create your own customizations?** Try `\u002Fxray extend`\n```\n\n---\n\n## MODE: Today (`\u002Fxray today`)\n\n**Purpose:** ScreenPipe-powered analysis of the day's activity.\n\n### Prerequisites\n\nFirst check if ScreenPipe is running. If not:\n\n```markdown\n## 🔬 X-Ray: Today\n\nScreenPipe isn't running, so I can't analyze your full day.\n\n**To enable day analysis:**\n1. Start ScreenPipe: Run `screenpipe` in terminal\n2. It records your screen activity locally (nothing sent to cloud)\n3. Come back and run `\u002Fxray today` again\n\n**What you'll get:**\n- Time breakdown by app\n- Activity patterns\n- \"Moments you missed\" where Amp could have helped\n```\n\n### If ScreenPipe is running\n\nQuery ScreenPipe for today's activity and present:\n\n```markdown\n## 🔬 X-Ray: Today\n\n[Activity summary from ScreenPipe]\n\n---\n\n### 🎓 How This Works\n\n**What you just saw was ScreenPipe integration.** Here's what's happening:\n\n1. **ScreenPipe records your screen** - Every few seconds, captures what's visible\n2. **OCR extracts text** - It can \"read\" what's on screen\n3. **A local database stores it** - SQLite on YOUR machine, not the cloud\n4. **I query that database** - Via ScreenPipe tools\n\n**Privacy model:**\n- Everything stays on YOUR computer\n- No data sent to external servers\n- You control retention\n\n**What this enables:**\n- \"What was I doing at 2pm?\" → I can look it up\n- \"How much time in Slack?\" → I can calculate\n- \"When did I see that error?\" → I can search\n\n---\n\n[Continue with activity analysis, patterns, and suggestions]\n```\n\n---\n\n## MODE: How to Extend (`\u002Fxray extend`)\n\n**Purpose:** Practical guide to customization.\n\n### Output Format\n\n```markdown\n## 🛠️ How to Extend Amp\n\nNow that you understand how it works, here's how to make it your own.\n\n---\n\n### Level 1: Edit CLAUDE.md (5 minutes)\n\n**The simplest customization.** Add your own instructions:\n\n1. Open `CLAUDE.md` in any editor\n2. Find the `USER_EXTENSIONS` section\n3. Add your rules:\n\n```markdown\n## USER_EXTENSIONS_START\n\n### My Preferences\n- I'm a morning person — suggest important work before noon\n- \"The board\" always means my company's board of directors\n- Never suggest scheduling meetings on Wednesdays (my focus day)\n\n## USER_EXTENSIONS_END\n```\n\n**This survives updates.** When you run `\u002Famp-update`, this block is preserved.\n\n---\n\n### Level 2: Create a Skill (30 minutes)\n\n**Make a reusable workflow:**\n\n1. Create folder: `.claude\u002Fskills\u002Fmy-skill\u002F`\n2. Create file: `SKILL.md`\n\n```markdown\n---\nname: my-skill\ndescription: What it does in one line\n---\n\n# My Custom Skill\n\n## When to Use\n[Trigger conditions]\n\n## Steps\n[Instructions for Claude to follow]\n\n## Output Format\n[What the result should look like]\n```\n\n3. Use it: Just type `\u002Fmy-skill`\n\n**Example ideas:**\n- `\u002Fstandup-prep` - Generate daily standup update\n- `\u002Fclient-brief` - Create client meeting brief\n- `\u002Fweekly-email` - Draft weekly summary email\n\n---\n\n### Level 3: Create a Hook (1 hour)\n\n**Inject context at every session start:**\n\n1. Create a script that outputs the context you want\n2. The output gets injected into the AI's context window\n3. Now every conversation starts with that knowledge\n\n**Example ideas:**\n- Inject today's calendar at start\n- Load current sprint goals\n- Show recent Git commits\n\n---\n\n### Level 4: Create an MCP Server (Half day)\n\n**Give Claude new capabilities:**\n\nRun `\u002Fcreate-mcp` for guided setup.\n\n**Example ideas:**\n- Todoist MCP - Sync tasks with Todoist\n- Notion MCP - Read\u002Fwrite to Notion\n- Weather MCP - Get current conditions\n\n---\n\n### The Extension Philosophy\n\n**Amp is scaffolding, not a finished product.** It's a starting point you customize.\n\nThe goal isn't to use Amp \"as is\" forever — it's to understand it well \nenough to make it truly yours.\n\n**Question for you:** What do you wish the AI knew at the start of \nevery session? That's your first extension.\n```\n\n---\n\n## Usage Tracking\n\nAfter running `\u002Fxray`, silently update `System\u002Fusage_log.md`:\n- Check `X-ray transparency (\u002Fxray)` in the discovery section\n\nIf user runs specific educational modes, update the AI Education Progress section:\n- `\u002Fxray ai` → Check \"Understands context windows\", \"Understands tokens\"\n- `\u002Fxray amp` → Check \"Understands system prompts\", \"Understands tools\u002FMCPs\"\n- `\u002Fxray boot` → Check \"Understands session hooks\"\n- `\u002Fxray extend` → Check \"Understands vault architecture\"\n\n---\n\n## Tone Guidelines\n\n- **Concrete over abstract** — Use specific examples from their conversation\n- **Show don't tell** — Point to actual files, tools, actions\n- **Empowering not impressive** — Goal is their understanding, not your cleverness\n- **Inviting curiosity** — End with pathways to learn more\n\n---\n\n## Track Usage (Silent)\n\nUpdate `System\u002Fusage_log.md` to mark AI transparency education as used.\n\n**Analytics (Silent):**\n\nCall `track_event` with event_name `xray_used` and properties:\n- `mode` (context\u002Ftokens\u002Fprompts\u002Ftools\u002Fhooks\u002Fvault\u002Fextend)\n\nThis only fires if the user has opted into analytics. No action needed if it returns \"analytics_disabled\".\n","# \u002Fxray - The AI Education Experience\n\n**Default behavior:** Explain what just happened in THIS conversation — the specific tools, files, and context that were used, with educational explanations of WHY.\n\n**With qualifiers:** Deep dive into specific topics (AI fundamentals, Amp architecture, how to extend).\n\n---\n\n## Philosophy\n\nThe best way to learn is by examining what just happened, not abstract concepts. When you run `\u002Fxray`:\n\n1. **Show the work** - What tools ran, what files were read\u002Fwritten\n2. **Explain why** - Why each action was necessary\n3. **Connect to concepts** - Link specific actions to how AI\u002FAmp works\n4. **Empower extension** - \"You could customize this by...\"\n\n**The outcome:** Users learn AI by examining their own conversations, then go deeper if curious.\n\n**Tone:** Like a senior engineer doing a code review of what just happened. \"Here's what I did and why.\"\n\n---\n\n## Step 1: Determine Mode\n\n### Default (no arguments): \"This Conversation\"\n\nWhen user just says `\u002Fxray` with no qualifier:\n→ **Immediately explain what happened in this conversation**\n→ Don't show a menu, just show the work\n\n### With qualifiers: Specific topics\n\n| Command | What it does |\n|---------|--------------|\n| `\u002Fxray` | (default) Explain what just happened in this chat |\n| `\u002Fxray ai` or `\u002Fxray fundamentals` | First principles: context windows, tokens, statelessness |\n| `\u002Fxray amp` or `\u002Fxray architecture` | How Amp works: CLAUDE.md, hooks, MCPs, skills |\n| `\u002Fxray boot` or `\u002Fxray session start` | The session startup sequence |\n| `\u002Fxray today` | ScreenPipe analysis of your day |\n| `\u002Fxray extend` or `\u002Fxray customize` | How to customize and build on Amp yourself |\n\n---\n\n## DEFAULT MODE: This Conversation (Show Your Work)\n\n**This is the primary mode.** When user runs `\u002Fxray`, explain what happened.\n\n### Step 1: Analyze the conversation\n\nLook at the current conversation and identify:\n1. What files were read\n2. What files were written\u002Fmodified\n3. What tools\u002FMCPs were used\n4. What context was loaded at session start (check for `\u003Cprevious_session_memories>` etc.)\n5. Key decision points in the AI's reasoning\n\n### Step 2: Present as educational walkthrough\n\nOutput format:\n\n```markdown\n## 🔬 X-Ray: What Just Happened\n\nLet me explain what happened under the hood in our conversation.\n\n---\n\n### 📖 The Story of This Chat\n\nHere's what I did, step by step:\n\n**1. Session Started**\nBefore you typed anything, the system loaded:\n- `CLAUDE.md` (~X tokens) — My personality, your preferences, available tools\n- Session memories from earlier today — [list what was injected]\n- [Any other context visible from hooks]\n\n**Why this matters:** Without those session memories, I'd have no idea about \n[specific context]. That came from a **session hook** that runs before you type.\n\n**2. You Asked: \"[summarize their first message]\"**\n\nTo respond, I:\n- Read `[file 1]` — Because [reason]\n- Read `[file 2]` — To check [what]\n- Used [tool\u002FMCP] — To [action]\n\n**Why these files?** Your CLAUDE.md tells me where to look. It says [quote relevant \ninstruction]. Without those instructions, I wouldn't know your file structure.\n\n**3. [Continue for each major exchange in the conversation]**\n\n---\n\n### 🔧 Tools Used This Session\n\n| Tool | Times Used | What For |\n|------|------------|----------|\n| `read` | X | Reading vault files |\n| `write` | X | Creating new content |\n| `edit` | X | Updating existing files |\n| `bash` | X | Running commands |\n| [other tools] | X | [purpose] |\n\n**Why tools matter:** Without tools, I can only generate text. Tools let me \nactually DO things — read your files, write new ones, check your calendar, etc.\n\n---\n\n### 💡 Key Insight: The Context Window\n\nEverything I \"know\" right now is what's in the **context window**:\n\n```\n┌─────────────────────────────────────────────────┐\n│  What I can see (the context window)            │\n├─────────────────────────────────────────────────┤\n│ ✓ CLAUDE.md (loaded at start)                   │\n│ ✓ Session memories (injected by hooks)          │\n│ ✓ Our full conversation so far                  │\n│ ✓ Every file I've read this session             │\n│ ✓ Every tool result I've received               │\n├─────────────────────────────────────────────────┤\n│ ✗ Files I haven't read                          │\n│ ✗ Previous sessions (unless summarized)         │\n│ ✗ Anything not loaded into context              │\n└─────────────────────────────────────────────────┘\n```\n\nIf something isn't in this window, I literally don't know it exists.\n\n---\n\n### 📁 Files Touched This Session\n\n**Read (loaded into my context):**\n- `[file]` — [why it was needed]\n- [continue for each file]\n\n**Written\u002FModified:**\n- `[file]` — [what was created\u002Fchanged]\n- [continue for each file]\n\n**Why this matters:** Every file I read uses \"tokens\" (space in the context \nwindow). I try to read only what's relevant, not everything.\n\n---\n\n### 🎓 Concepts You Just Saw in Action\n\n| What Happened | The Underlying Concept |\n|---------------|------------------------|\n| I knew about earlier conversations | **Session hooks** inject context at start |\n| I read CLAUDE.md first | **System prompt** loads before anything |\n| I created files with tools | **Tools** let AI take action, not just talk |\n| I followed your file conventions | **CLAUDE.md instructions** define your structure |\n| I didn't know about [X] | **Context limitation** — only know what's loaded |\n\n---\n\n### 🔮 What You Could Customize\n\nBased on what just happened:\n\n**1. Add to CLAUDE.md**\nIf you want me to always [behavior], add it to your `USER_EXTENSIONS` block.\n\n**2. Create a Skill**\nIf you do [this workflow] often, create a skill to automate it.\n\n**3. Add a Session Hook**\nIf you always want [context] loaded at start, write a hook that injects it.\n\n---\n\n### 🎓 Want to Go Deeper?\n\n- `\u002Fxray ai` — First principles: context windows, tokens, statelessness\n- `\u002Fxray amp` — Full architecture: CLAUDE.md, hooks, MCPs, skills, vault\n- `\u002Fxray boot` — The session startup sequence in detail\n- `\u002Fxray extend` — Step-by-step guide to customizing\n\nOr ask anything: \"How do hooks work?\", \"What's an MCP?\", \"How can I add my own tools?\"\n```\n\n---\n\n## MODE: AI Fundamentals (`\u002Fxray ai`)\n\n**Purpose:** Teach the core concepts that make everything else make sense.\n\n### Output Format\n\n```markdown\n## 🧠 How AI Actually Works\n\nLet me explain the core concepts that underpin everything Amp does.\n\n---\n\n### The Big Truth: AI Has No Memory\n\n**The fundamental problem:** Every time you start a new chat, Claude has amnesia. \nIt doesn't remember yesterday's conversation. It doesn't know your name, your \nprojects, or what you discussed last week.\n\nThis is true for ALL LLM-based AI — ChatGPT, Claude, Gemini, all of them.\n\n**Why?** LLMs are \"stateless.\" They process text in, text out, with no permanent \nstorage between sessions. Each conversation starts from zero.\n\n**So how does Amp remember things?** We engineer around the limitation. \nThat's what Amp is — clever engineering to give a forgetful AI the *appearance* \nof memory, context, and continuity.\n\n---\n\n### Concept 1: The Context Window\n\nWhen you chat with an AI, there's an invisible \"window\" of text the AI can see:\n\n```\n┌─────────────────────────────────────────────────┐\n│              THE CONTEXT WINDOW                 │\n│  (Everything the AI knows about THIS chat)      │\n├─────────────────────────────────────────────────┤\n│ • System prompt (CLAUDE.md)                     │\n│ • Injected context (session memories, etc.)     │\n│ • Our conversation so far                       │\n│ • Any files I've read                           │\n│ • Tool results (MCP responses)                  │\n└─────────────────────────────────────────────────┘\n          ↓\n     AI generates response based on ALL of this\n```\n\n**Key insight:** The AI can only work with what's IN the context window. \nIf information isn't in there, the AI doesn't know it exists.\n\n**Amp's job:** Get the RIGHT information INTO that window at the RIGHT time.\n\n---\n\n### Concept 2: Tokens (The Currency)\n\nContext windows are measured in \"tokens\" (roughly 4 characters = 1 token).\n\n- Claude's window: ~200,000 tokens (~150,000 words)\n- Your CLAUDE.md: ~4,500 tokens\n- A typical person page: ~500 tokens\n\n**Why it matters:** There's a budget. Load too much context and you hit limits \nor slow things down. Load too little and the AI lacks crucial information.\n\n**Amp's approach:** Selective loading. Don't dump everything — load what's \n*relevant* to what you're asking about.\n\n---\n\n### Concept 3: System Prompts (The Personality)\n\nBefore your first message, there's already text in the context window: \nthe **system prompt**. For Amp, that's `CLAUDE.md`.\n\n```\n┌─────────────────────────────────────────────────┐\n│ CLAUDE.md (loaded before you type anything)     │\n├─────────────────────────────────────────────────┤\n│ \"You are Amp, a personal knowledge assistant\"   │\n│ \"User's pillars: [your pillars]\"                │\n│ \"When tasks are mentioned, use Work MCP\"        │\n│ \"Person pages are in 05-Areas\u002FPeople\u002F\"          │\n│ ... (hundreds more lines of context)            │\n└─────────────────────────────────────────────────┘\n```\n\n**The power move:** Everything in CLAUDE.md influences EVERY response. \nIt's like setting the AI's personality and knowledge before you say hello.\n\n---\n\n### Concept 4: Tools (How AI Does Things)\n\nClaude can't *do* things by default — it can only generate text. But with \n**tools**, it can take actions:\n\n- Read files from your computer\n- Write new files\n- Run commands\n- Query databases\n- Call APIs\n\nIn Amp, tools come from **MCP servers** (Model Context Protocol). Each MCP \ngives Claude new capabilities.\n\n**Without tools:** Claude can only talk.\n**With tools:** Claude can take action on your behalf.\n\n---\n\n### Concept 5: The Conversation Loop\n\nHere's what happens every time you send a message:\n\n```\n1. Your message gets added to context window\n                    ↓\n2. Claude sees: system prompt + history + your message\n                    ↓\n3. Claude decides: should I use a tool?\n   - If yes → Tool runs → Result added to context → Back to step 3\n   - If no → Generate response\n                    ↓\n4. Response appears (and gets added to context for next turn)\n                    ↓\n5. Wait for your next message → Repeat from step 1\n```\n\n**Key insight:** Each turn, Claude sees MORE context (the history grows). \nLong conversations = more context = closer to limits.\n\n---\n\n### So What?\n\nUnderstanding these fundamentals helps you:\n\n1. **Know why Amp loads things at session start** — to get context into the window\n2. **Understand why some info is \"missing\"** — it wasn't loaded into context\n3. **See opportunities to extend** — what else could be loaded? what tools could help?\n\n---\n\n**Want to see Amp's specific architecture?** Try `\u002Fxray amp`\n```\n\n---\n\n## MODE: Amp Architecture (`\u002Fxray amp`)\n\n**Purpose:** Explain the specific building blocks that make Amp work.\n\n### Output Format\n\n```markdown\n## 🏗️ How Amp Works: The Architecture\n\nNow that you understand AI fundamentals, here's how Amp engineers around \nthe limitations to create something useful.\n\n---\n\n### The Core Insight\n\n**Problem:** Claude forgets everything between sessions.  \n**Solution:** Store everything in FILES and reload context intelligently.\n\n```\n┌─────────────────────────────────────────────────┐\n│                YOUR VAULT                        │\n│  (Plain Markdown files YOU own)                 │\n├─────────────────────────────────────────────────┤\n│ • CLAUDE.md - Core context & personality        │\n│ • 05-Areas\u002FPeople\u002F - Relationship memory        │\n│ • 03-Tasks\u002FTasks.md - Work in progress          │\n│ • System\u002FSession_Memory\u002F - Past conversations   │\n│ • System\u002FWork_In_Progress.md - Active projects  │\n└─────────────────────────────────────────────────┘\n            ↓ loaded at session start ↓\n         ┌─────────────────────────────┐\n         │      AI Context Window      │\n         └─────────────────────────────┘\n```\n\n**The magic:** Your knowledge lives in files. AI reads files. \nTherefore, AI can \"remember\" by reading what was written before.\n\n---\n\n### Building Block 1: CLAUDE.md (The Brain)\n\nThe system prompt that defines who Amp is and how it behaves.\n\n**Location:** `\u002FCLAUDE.md` (root of your vault)\n\n**Contains:**\n- Your user profile (name, role, company)\n- Your strategic pillars\n- How Amp should behave\n- What skills are available\n- References to other docs\n\n**Why it matters:** This loads FIRST, before anything else. It shapes every \nresponse. Edit this, and you change how Amp thinks.\n\n**You can customize:**\n```markdown\n## USER_EXTENSIONS_START\n\u003C!-- Your personal additions go here -->\n## USER_EXTENSIONS_END\n```\n\n---\n\n### Building Block 2: Session Hooks (The Boot Sequence)\n\nWhen you start a chat, things happen BEFORE you type anything.\n\n**What hooks do:**\n- Inject session memories from recent conversations\n- Surface relevant context (person info, project status)\n- Pre-load data you'll likely need\n\n**Current hooks in Amp:**\n- Session memories injection\n- Person context surfacing\n- Company context surfacing\n\n**The power:** You can ADD hooks. Want weather loaded at session start? \nYour calendar? Write a hook, and it's there every session.\n\n---\n\n### Building Block 3: MCP Servers (The Toolbox)\n\nMCP = Model Context Protocol. It's how Claude gets superpowers.\n\nEach MCP server gives Claude new abilities:\n\n| Server | What It Does |\n|--------|--------------|\n| **Work MCP** | Task management with deduplication and pillar alignment |\n| **Calendar MCP** | Reads your Apple Calendar |\n| **Granola MCP** | Searches meeting transcripts |\n| **Career MCP** | Career evidence tracking and analysis |\n\n**Why MCPs instead of just reading files?**\n- **Speed:** Pre-process data instead of parsing raw files\n- **Logic:** Implement business rules (like \"can't have 4 P0 tasks\")\n- **Integration:** Connect to external systems\n\n---\n\n### Building Block 4: Skills (The Playbook)\n\nSkills are structured instructions for multi-step workflows.\n\n**Location:** `.claude\u002Fskills\u002F[skill-name]\u002FSKILL.md`\n\n**Examples:**\n- `\u002Fdaily-plan` - Check calendar, review tasks, generate plan\n- `\u002Fmeeting-prep` - Load person context, surface recent discussions\n- `\u002Fxray` - This skill you're running right now!\n\n**Why skills?**\n- **Consistency:** Same workflow every time\n- **Complexity:** Handle multi-step processes\n- **Reusability:** Build once, use forever\n\n---\n\n### Building Block 5: The Vault (The Memory)\n\nEverything persists to Markdown files:\n\n```\nYour Vault\u002F\n├── 00-Inbox\u002F           ← Capture zone\n├── 03-Tasks\u002F           ← Active tasks\n├── 04-Projects\u002F        ← Time-bound work\n├── 05-Areas\u002F           ← Ongoing responsibilities\n│   ├── People\u002F         ← Person pages\n│   └── Companies\u002F      ← Company profiles\n├── 06-Resources\u002F       ← Reference material\n└── System\u002F             ← Amp configuration\n```\n\n**Why plain Markdown?**\n- You own it forever (no vendor lock-in)\n- Works anywhere (Obsidian, VS Code, any editor)\n- AI can read and write it\n- Version control with Git if you want\n\n---\n\n### How It All Connects\n\n```\nSession starts\n     ↓\nHooks fire (inject context)\n     ↓\nCLAUDE.md loads (personality + instructions)\n     ↓\nYou type your message\n     ↓\nClaude reads relevant files (on-demand)\n     ↓\nClaude uses MCPs (if needed)\n     ↓\nResponse generated + files updated\n     ↓\nRepeat for each message\n```\n\n---\n\n### Key Insight: It's All Text\n\nThere's no magic database. No proprietary format. No cloud service.\n\n- CLAUDE.md is text\n- Skills are text\n- Your notes are text\n- Session memories are text\n\nClaude reads text, processes it, generates text, writes text.\n\n**Everything you're experiencing is text transformation.**\n\nUnderstanding this is liberating — you can read, edit, and extend anything.\n\n---\n\n**Want to see the session startup in detail?** Try `\u002Fxray boot`\n**Ready to customize?** Try `\u002Fxray extend`\n```\n\n---\n\n## MODE: Session Boot Sequence (`\u002Fxray boot`)\n\n**Purpose:** Detailed walkthrough of what happens when a session starts.\n\n### Output Format\n\n```markdown\n## 🚀 What Happens When You Start a Chat\n\nLet me walk you through exactly what happens from the moment you open \na new chat to when you type your first message.\n\n---\n\n### The Boot Sequence (In Order)\n\n#### Step 1: Environment Detection\n```\nIs this Claude Desktop? Claude Code? Cursor? PI?\n     ↓\nEach environment has slightly different capabilities\n```\n\n#### Step 2: Settings Load\n```\nRead configuration files\n     ↓\nThis defines which hooks to run and which MCPs are available\n```\n\n#### Step 3: Hooks Fire (CRITICAL)\n```\nPre-session hooks run in order:\n     ↓\n┌─────────────────────────────────────────────┐\n│ Session memory injection                    │\n│ → Reads System\u002FSession_Memory\u002F[date].md     │\n│ → Injects summaries of recent conversations │\n│ → AI \"remembers\" what you discussed before  │\n└─────────────────────────────────────────────┘\n     ↓\n┌─────────────────────────────────────────────┐\n│ Context injectors                           │\n│ → Person context ready when you mention     │\n│   names                                     │\n│ → Company context ready when relevant       │\n└─────────────────────────────────────────────┘\n     ↓\n┌─────────────────────────────────────────────┐\n│ (Your custom hooks could go here)           │\n│ → Whatever YOU want loaded at start         │\n└─────────────────────────────────────────────┘\n```\n\n**This is where the magic happens!** Hooks let you inject ANY context \nbefore the conversation starts.\n\n#### Step 4: CLAUDE.md Loads\n```\nThe entire CLAUDE.md file enters the context window:\n     ↓\n• Your profile (name, role, company)\n• Your strategic pillars  \n• Behavior guidelines\n• Reference to skills and tools\n• User extensions (YOUR customizations)\n```\n\n#### Step 5: MCPs Connect\n```\nMCP servers become available:\n     ↓\n• Calendar MCP → Ready to check schedule\n• Work MCP → Ready to manage tasks\n• Granola MCP → Ready to search meetings\n```\n\n#### Step 6: Ready for Input\n```\nNOW you can type your first message.\n\nThe AI already has:\n• Your personality and preferences (CLAUDE.md)\n• Recent conversation summaries (from hooks)\n• Relevant context (from hooks)\n• Access to tools (MCPs)\n```\n\n---\n\n### What This Means for You\n\n**The AI doesn't start from zero.** By the time you type your first message, \nClaude already knows who you are, what you've been working on, and what \ntools are available.\n\n**The boot sequence IS the memory system.** It's how a stateless AI \n\"remembers\" who you are.\n\n---\n\n### 💡 Customization Opportunity\n\n**What else could be loaded at session start?**\n\nIdeas for hooks:\n1. Today's calendar events\n2. Current sprint goals\n3. Recent Git commits\n4. Weather for outdoor planning\n5. Your daily intention\n\n**The pattern:** Anything you want the AI to \"know\" at the start of every \nsession can be injected via a hook.\n\n---\n\n**Ready to create your own customizations?** Try `\u002Fxray extend`\n```\n\n---\n\n## MODE: Today (`\u002Fxray today`)\n\n**Purpose:** ScreenPipe-powered analysis of the day's activity.\n\n### Prerequisites\n\nFirst check if ScreenPipe is running. If not:\n\n```markdown\n## 🔬 X-Ray: Today\n\nScreenPipe isn't running, so I can't analyze your full day.\n\n**To enable day analysis:**\n1. Start ScreenPipe: Run `screenpipe` in terminal\n2. It records your screen activity locally (nothing sent to cloud)\n3. Come back and run `\u002Fxray today` again\n\n**What you'll get:**\n- Time breakdown by app\n- Activity patterns\n- \"Moments you missed\" where Amp could have helped\n```\n\n### If ScreenPipe is running\n\nQuery ScreenPipe for today's activity and present:\n\n```markdown\n## 🔬 X-Ray: Today\n\n[Activity summary from ScreenPipe]\n\n---\n\n### 🎓 How This Works\n\n**What you just saw was ScreenPipe integration.** Here's what's happening:\n\n1. **ScreenPipe records your screen** - Every few seconds, captures what's visible\n2. **OCR extracts text** - It can \"read\" what's on screen\n3. **A local database stores it** - SQLite on YOUR machine, not the cloud\n4. **I query that database** - Via ScreenPipe tools\n\n**Privacy model:**\n- Everything stays on YOUR computer\n- No data sent to external servers\n- You control retention\n\n**What this enables:**\n- \"What was I doing at 2pm?\" → I can look it up\n- \"How much time in Slack?\" → I can calculate\n- \"When did I see that error?\" → I can search\n\n---\n\n[Continue with activity analysis, patterns, and suggestions]\n```\n\n---\n\n## MODE: How to Extend (`\u002Fxray extend`)\n\n**Purpose:** Practical guide to customization.\n\n### Output Format\n\n```markdown\n## 🛠️ How to Extend Amp\n\nNow that you understand how it works, here's how to make it your own.\n\n---\n\n### Level 1: Edit CLAUDE.md (5 minutes)\n\n**The simplest customization.** Add your own instructions:\n\n1. Open `CLAUDE.md` in any editor\n2. Find the `USER_EXTENSIONS` section\n3. Add your rules:\n\n```markdown\n## USER_EXTENSIONS_START\n\n### My Preferences\n- I'm a morning person — suggest important work before noon\n- \"The board\" always means my company's board of directors\n- Never suggest scheduling meetings on Wednesdays (my focus day)\n\n## USER_EXTENSIONS_END\n```\n\n**This survives updates.** When you run `\u002Famp-update`, this block is preserved.\n\n---\n\n### Level 2: Create a Skill (30 minutes)\n\n**Make a reusable workflow:**\n\n1. Create folder: `.claude\u002Fskills\u002Fmy-skill\u002F`\n2. Create file: `SKILL.md`\n\n```markdown\n---\nname: my-skill\ndescription: What it does in one line\n---\n\n# My Custom Skill\n\n## When to Use\n[Trigger conditions]\n\n## Steps\n[Instructions for Claude to follow]\n\n## Output Format\n[What the result should look like]\n```\n\n3. Use it: Just type `\u002Fmy-skill`\n\n**Example ideas:**\n- `\u002Fstandup-prep` - Generate daily standup update\n- `\u002Fclient-brief` - Create client meeting brief\n- `\u002Fweekly-email` - Draft weekly summary email\n\n---\n\n### Level 3: Create a Hook (1 hour)\n\n**Inject context at every session start:**\n\n1. Create a script that outputs the context you want\n2. The output gets injected into the AI's context window\n3. Now every conversation starts with that knowledge\n\n**Example ideas:**\n- Inject today's calendar at start\n- Load current sprint goals\n- Show recent Git commits\n\n---\n\n### Level 4: Create an MCP Server (Half day)\n\n**Give Claude new capabilities:**\n\nRun `\u002Fcreate-mcp` for guided setup.\n\n**Example ideas:**\n- Todoist MCP - Sync tasks with Todoist\n- Notion MCP - Read\u002Fwrite to Notion\n- Weather MCP - Get current conditions\n\n---\n\n### The Extension Philosophy\n\n**Amp is scaffolding, not a finished product.** It's a starting point you customize.\n\nThe goal isn't to use Amp \"as is\" forever — it's to understand it well \nenough to make it truly yours.\n\n**Question for you:** What do you wish the AI knew at the start of \nevery session? That's your first extension.\n```\n\n---\n\n## Usage Tracking\n\nAfter running `\u002Fxray`, silently update `System\u002Fusage_log.md`:\n- Check `X-ray transparency (\u002Fxray)` in the discovery section\n\nIf user runs specific educational modes, update the AI Education Progress section:\n- `\u002Fxray ai` → Check \"Understands context windows\", \"Understands tokens\"\n- `\u002Fxray amp` → Check \"Understands system prompts\", \"Understands tools\u002FMCPs\"\n- `\u002Fxray boot` → Check \"Understands session hooks\"\n- `\u002Fxray extend` → Check \"Understands vault architecture\"\n\n---\n\n## Tone Guidelines\n\n- **Concrete over abstract** — Use specific examples from their conversation\n- **Show don't tell** — Point to actual files, tools, actions\n- **Empowering not impressive** — Goal is their understanding, not your cleverness\n- **Inviting curiosity** — End with pathways to learn more\n\n---\n\n## Track Usage (Silent)\n\nUpdate `System\u002Fusage_log.md` to mark AI transparency education as used.\n\n**Analytics (Silent):**\n\nCall `track_event` with event_name `xray_used` and properties:\n- `mode` (context\u002Ftokens\u002Fprompts\u002Ftools\u002Fhooks\u002Fvault\u002Fextend)\n\nThis only fires if the user has opted into analytics. No action needed if it returns \"analytics_disabled\".\n",{"name":5,"description":7},"main","6a447b9ff35eed7bda9e12399bcd373dd31e0a53",".claude\u002Fskills\u002Fxray","https:\u002F\u002Fgithub.com\u002Fhemory\u002Famp","https:\u002F\u002Fgithub.com\u002Fhemory","https:\u002F\u002Fwww.heminthehat.com\u002F","The People Developer | L&D Program Manager | Software Engineer","2026-06-09 00:20:04",{"source":8,"skillName":5,"commands":33,"prompt":43,"zipAvailable":44},[34,37,40],{"tool":35,"command":36},"npx","npx skills add hemory\u002Famp --skill xray",{"tool":38,"command":39},"bunx","bunx skills add hemory\u002Famp --skill xray",{"tool":41,"command":42},"pnpm","pnpm dlx skills add hemory\u002Famp --skill xray","安装技能，技能地址：https:\u002F\u002Fgithub.com\u002Fhemory\u002Famp\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fxray\u002FSKILL.md",true,[],[47,54,61,68,75,82],{"skillNo":48,"name":49,"nameZh":6,"description":50,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":51,"categories":52,"keywords":53,"publishedAt":19},"01M76W5GGTP0YHZQQJ4G2ZPSZE","zoom-setup","Connect Zoom to Amp for meeting recordings, scheduling, and transcript context","https:\u002F\u002Fgithub.com\u002Fhemory\u002Famp\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fzoom-setup\u002FSKILL.md",[],[],{"skillNo":55,"name":56,"nameZh":6,"description":57,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":58,"categories":59,"keywords":60,"publishedAt":19},"01M76W582J6PD3TTXZXFA2HVQJ","week-review","Review week's progress with concrete accomplishments (not fake percentages), pattern detection, and goal tracking.","https:\u002F\u002Fgithub.com\u002Fhemory\u002Famp\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fweek-review\u002FSKILL.md",[],[],{"skillNo":62,"name":63,"nameZh":6,"description":64,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":65,"categories":66,"keywords":67,"publishedAt":19},"01M76W54EY8D8Y55DEPYPJ9YBV","week-plan","Set weekly priorities and plan the week ahead with intelligent suggestions based on goals, calendar shape, and task effort.","https:\u002F\u002Fgithub.com\u002Fhemory\u002Famp\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Fweek-plan\u002FSKILL.md",[],[],{"skillNo":69,"name":70,"nameZh":6,"description":71,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":72,"categories":73,"keywords":74,"publishedAt":19},"01M76W527EJ43YXBN0CE9SRAYT","triage","Strategically route orphaned files and extract scattered tasks","https:\u002F\u002Fgithub.com\u002Fhemory\u002Famp\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Ftriage\u002FSKILL.md",[],[],{"skillNo":76,"name":77,"nameZh":6,"description":78,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":79,"categories":80,"keywords":81,"publishedAt":19},"01M76W4ZYRDVHZ3SS8W5WZ53Q8","trello-setup","Connect Trello to Amp for visual Kanban task sync","https:\u002F\u002Fgithub.com\u002Fhemory\u002Famp\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Ftrello-setup\u002FSKILL.md",[],[],{"skillNo":83,"name":84,"nameZh":6,"description":85,"descriptionZh":6,"repositoryFullName":8,"ownerLogin":9,"ownerAvatarUrl":10,"ownerAccountNo":11,"ownerDisplayName":12,"repositoryNo":13,"starsCount":14,"licenseSpdxId":15,"licenseFileUrl":6,"sourceHtmlUrl":86,"categories":87,"keywords":88,"publishedAt":19},"01M76W4WXPFTAQT7KM8E65QVBE","todoist-setup","Connect Todoist to Amp for two-way task sync","https:\u002F\u002Fgithub.com\u002Fhemory\u002Famp\u002Fblob\u002Fmain\u002F.claude\u002Fskills\u002Ftodoist-setup\u002FSKILL.md",[],[]]