luna-agents 2.0.1

package/commands/ll-launch.md

@@ -0,0 +1,100 @@
+---
+name: ll-launch
+displayName: Luna Launch
+description: Full product launch sequence — deploy, monitor, announce, track metrics, auto-rollback if issues
+version: 1.0.0
+category: deployment
+agent: luna-deployment
+parameters:
+  - name: target
+    type: string
+    description: "Target: staging, production, preview"
+    required: true
+    prompt: true
+  - name: strategy
+    type: string
+    description: "Strategy: instant, canary (5%→25%→100%), blue-green, rolling"
+    required: false
+    default: canary
+mcp_servers:
+  - cloudflare
+  - git
+  - playwright
+  - zai-mcp-server
+  - image-compare
+  - accessibility-scanner
+  - fetch
+  - memory
+---
+
+# /launch — Ship with Confidence
+
+Not just deploy. A full launch sequence with visual verification, progressive rollout, real-time monitoring, and automatic rollback.
+
+## Launch Sequence
+
+```
+/launch production --strategy canary
+         │
+         ▼
+┌─── PRE-FLIGHT ────────────────────────────────┐
+│  git: verify clean branch, all tests pass      │
+│  playwright: full browser test suite            │
+│  accessibility-scanner: WCAG AA compliance      │
+│  image-compare: screenshot baseline captured    │
+│  fetch: verify all external dependencies up     │
+└────────────────────┬──────────────────────────┘
+                     ▼
+┌─── DEPLOY 5% CANARY ─────────────────────────┐
+│  cloudflare: deploy to edge, route 5% traffic  │
+│  wait 5 minutes                                │
+│  fetch: hit health endpoints                   │
+│  playwright: smoke test against canary URL      │
+│  zai-mcp: analyze canary screenshots vs baseline│
+│  CHECK: error rate < 0.1%? latency < p99?       │
+│  ├── PASS → promote to 25%                      │
+│  └── FAIL → auto-rollback, alert, report        │
+└────────────────────┬──────────────────────────┘
+                     ▼
+┌─── DEPLOY 25% ────────────────────────────────┐
+│  cloudflare: route 25% traffic                  │
+│  wait 10 minutes                               │
+│  Same verification as 5% step                  │
+│  CHECK: all signals green?                      │
+│  ├── PASS → promote to 100%                     │
+│  └── FAIL → auto-rollback to previous version   │
+└────────────────────┬──────────────────────────┘
+                     ▼
+┌─── FULL DEPLOY ───────────────────────────────┐
+│  cloudflare: route 100% traffic                 │
+│  playwright: full regression suite              │
+│  image-compare: diff vs pre-deploy baseline     │
+│  memory: store deployment record                │
+│  git: tag release                               │
+└────────────────────┬──────────────────────────┘
+                     ▼
+┌─── POST-LAUNCH ───────────────────────────────┐
+│  Generate deployment report                    │
+│  Store screenshots as new baseline             │
+│  Update changelog                              │
+│  memory: persist launch metrics for next time   │
+└───────────────────────────────────────────────┘
+```
+
+## Usage
+
+```bash
+/launch staging                                   # Instant deploy to staging
+/launch production                                # Canary deploy to production
+/launch production --strategy blue-green          # Blue-green swap
+/launch production --strategy rolling             # Rolling update
+/launch preview                                   # Deploy preview branch
+```
+
+## In Pipes
+
+```bash
+/pipe test >> rev >> sec >> launch production
+/pipe go *5 >> (test ~~ browser-test ~~ a11y) >> launch staging >> approve "Ship to prod?" >> launch production
+/pipe swarm "build feature" >> test >> launch staging >> visual-diff >> launch production
+```

package/commands/ll-learn.md

@@ -0,0 +1,87 @@
+---
+name: ll-learn
+displayName: Luna Learn
+description: Self-learning memory — Luna gets smarter with every task, remembering patterns, preferences, and successful approaches
+version: 1.0.0
+category: intelligence
+agent: luna-task-executor
+parameters:
+  - name: action
+    type: string
+    description: "Action: status (show what Luna knows), teach (add knowledge), forget (remove), export, import"
+    required: false
+    default: status
+mcp_servers:
+  - memory
+  - ruflo
+  - sequential-thinking
+---
+
+# /learn — Luna Gets Smarter Over Time
+
+Every command you run teaches Luna. Successful patterns are stored, failures are analyzed, and your preferences are remembered. Luna evolves with you.
+
+## The Learning Loop
+
+```
+You run a command → Luna executes → Outcome observed
+                                          │
+                    ┌─────────────────────┤
+                    ▼                     ▼
+              SUCCESS                 FAILURE
+              │                       │
+              ▼                       ▼
+        DISTILL pattern         ANALYZE root cause
+        Store in memory         Store anti-pattern
+              │                       │
+              └───────────┬───────────┘
+                          ▼
+                    MEMORY BANK
+                    │
+                    ├── Your code style preferences
+                    ├── Your architecture patterns
+                    ├── Your test strategy
+                    ├── Your deploy workflow
+                    ├── Your review standards
+                    ├── Common bugs in your codebase
+                    ├── Successful fix patterns
+                    └── Failed approaches to avoid
+                          │
+                          ▼
+                    NEXT COMMAND
+                    Luna applies learned patterns
+                    Avoids known failures
+                    Matches your style
+```
+
+## What Luna Learns
+
+| Category | Examples |
+|----------|---------|
+| **Code Style** | You prefer functional over class components, named exports, barrel files |
+| **Architecture** | You use Zustand over Redux, Hono over Express, Zod for validation |
+| **Testing** | You write integration tests first, mock at boundaries, 90%+ coverage |
+| **Review** | You care about bundle size, hate magic numbers, enforce 200-line limit |
+| **Deploy** | You use canary deploys, always run visual QA before production |
+| **Naming** | camelCase for functions, PascalCase for components, kebab-case for files |
+| **Failures** | "Don't use SPA mode with serve for static sites" (learned from this session!) |
+
+## Usage
+
+```bash
+/learn                     # Show what Luna knows about you
+/learn status              # Same — full knowledge dashboard
+/learn teach "Always use Bun over npm for this project"
+/learn teach "Our API uses snake_case, not camelCase"
+/learn forget "old pattern about Redux"
+/learn export              # Export knowledge as JSON
+/learn import ./team.json  # Import team's shared knowledge
+```
+
+## In Pipes
+
+```bash
+/pipe learn >> go *5                    # Apply learned patterns then code
+/pipe go *5 >> test >> learn            # Code, test, then distill learnings
+/pipe learn export >> share team        # Share learnings with team
+```

package/commands/ll-leverage.md

@@ -0,0 +1,137 @@
+---
+name: ll-leverage
+displayName: Luna Leverage Open Source
+description: Scan open-source repos, extract reusable patterns/features, and generate integration plans for your project
+version: 1.0.0
+category: analysis
+agent: luna-task-executor
+parameters:
+  - name: repos
+    type: string
+    description: "GitHub repo URLs (comma-separated) or 'user:username' to scan all repos"
+    required: true
+    prompt: true
+  - name: target
+    type: string
+    description: "Your project to integrate into (e.g., 'lunaos', 'coderailflow', or path)"
+    required: false
+    default: current
+  - name: depth
+    type: string
+    description: "Analysis depth: quick (README only), medium (+ key files), deep (full codebase)"
+    required: false
+    default: medium
+prerequisites: []
+---
+
+# Luna Leverage Open Source
+
+Scan any GitHub repo (or an entire user's repos), extract reusable features, patterns, and architecture decisions, then generate an integration plan for your project.
+
+## What This Command Does
+
+1. **Scan** — Fetches repo README, key source files, package.json, architecture
+2. **Extract** — Identifies: features, tech stack, design patterns, APIs, unique innovations
+3. **Compare** — Maps against your project's existing capabilities
+4. **Gap Analysis** — Finds what the repo has that your project doesn't
+5. **Integration Plan** — Generates specific code-level steps to adopt the best parts
+6. **Reference Doc** — Creates a markdown reference sheet for team knowledge
+
+## Usage
+
+```bash
+# Scan a single repo
+/leverage https://github.com/ruvnet/ruflo --target lunaos
+
+# Scan all repos from a user
+/leverage user:ruvnet --target lunaos --depth quick
+
+# Scan multiple repos
+/leverage https://github.com/ruvnet/ruflo,https://github.com/ruvnet/RuVector --target lunaos
+
+# Deep analysis with full codebase scan
+/leverage https://github.com/elie222/inbox-zero --depth deep
+
+# Just generate reference docs (no integration plan)
+/leverage https://github.com/karpathy/nanoGPT --depth quick
+```
+
+## Output
+
+```
+.luna/{project}/leverage/
+  {repo-name}/
+    reference.md          # What it is, features, stack, unique patterns
+    gap-analysis.md       # What it has that you don't
+    integration-plan.md   # Step-by-step adoption guide
+    key-files.md          # Most important source files + summaries
+  summary.md              # Cross-repo comparison matrix
+  adoption-roadmap.md     # Prioritized plan across all scanned repos
+```
+
+## Reference Doc Format
+
+For each scanned repo, generates:
+
+```markdown
+# {Repo Name}
+**URL**: {url}
+**Stars**: {count} | **Language**: {lang} | **License**: {license}
+
+## What It Does
+{2-3 sentence description}
+
+## Key Features
+- Feature 1: {description}
+- Feature 2: {description}
+
+## Tech Stack
+| Layer | Technology |
+|-------|-----------|
+| ...   | ...       |
+
+## Architecture Patterns Worth Adopting
+1. {Pattern}: {why it's good, how to apply}
+
+## Unique Innovations
+- {Innovation}: {what makes it special}
+
+## API / Integration Points
+- {endpoint/SDK}: {how to call it}
+
+## Relevance to {Your Project}
+- Direct competitor: Yes/No
+- Features to adopt: {list}
+- Estimated integration effort: {hours}
+```
+
+## How It Works
+
+### Quick Scan
+- Fetches README.md via GitHub API
+- Extracts: description, features, stack, stars
+- Generates reference doc only
+
+### Medium Scan (default)
+- Quick scan + reads key files (package.json, main entry, config)
+- Identifies architecture patterns
+- Generates gap analysis + integration plan
+
+### Deep Scan
+- Medium scan + clones repo to temp dir
+- Analyzes full source tree (respects .gitignore)
+- Maps every feature to your project's equivalent
+- Generates code-level integration steps
+
+## In Pipes
+
+```bash
+# Scan competitor, then integrate best features
+/pipe leverage https://github.com/competitor/app >> fix
+
+# Scan, plan, execute
+/pipe leverage https://github.com/cool/project >> product-map >> go
+
+# Research phase: scan multiple, then plan
+/pipe leverage user:ruvnet --depth quick >> context-pack >> plan
+```

package/commands/ll-local-llm.md

@@ -0,0 +1,131 @@
+---
+name: ll-local-llm
+displayName: Luna Local LLM
+description: Run AI agents locally using llamafile — zero cost, fully offline, OpenAI-compatible API
+version: 1.0.0
+category: infrastructure
+agent: luna-task-executor
+parameters:
+  - name: action
+    type: string
+    description: "Action: start (launch llamafile server), stop (shut down), status (check running)"
+    required: true
+    prompt: true
+  - name: model
+    type: string
+    description: "Model to run (default: qwen3.5-0.8b)"
+    required: false
+    default: qwen3.5-0.8b
+mcp_servers:
+  - memory
+  - git
+prerequisites:
+  - name: llamafile
+    check: "which llamafile || ls ~/.local/bin/llamafile"
+    install: "curl -fLo llamafile https://github.com/mozilla-ai/llamafile/releases/latest/download/llamafile && chmod +x llamafile && mv llamafile ~/.local/bin/"
+    optional: false
+---
+
+# /local-llm — Zero-Cost Local AI with llamafile
+
+Run AI agents entirely offline using llamafile. Downloads and runs a single binary that exposes an OpenAI-compatible API at `localhost:8080`. Use Luna CLI's `--local` flag to route all agent calls through llamafile instead of Claw Gateway.
+
+## What It Does
+
+```
+/local-llm start
+    │
+    ├── CHECK: llamafile binary available?
+    │   ├── YES → Continue
+    │   └── NO  → Download from GitHub releases
+    │
+    ├── DOWNLOAD MODEL (if not cached)
+    │   ├── qwen3.5-0.8b (default, 800 MB, fast)
+    │   ├── llama-3.1-8b (8 GB, balanced)
+    │   ├── mistral-7b (7 GB, general purpose)
+    │   └── codellama-13b (13 GB, code-focused)
+    │
+    ├── START SERVER
+    │   ├── Launch llamafile on localhost:8080
+    │   ├── OpenAI-compatible /v1/chat/completions
+    │   ├── GPU auto-detection (Metal, CUDA, Vulkan)
+    │   └── Print status + connection URL
+    │
+    └── CONFIGURE LUNA
+        ├── Set LUNA_LOCAL=true
+        ├── Route --local flag to localhost:8080
+        └── All agents use local model
+```
+
+## How It Works
+
+1. **Single binary**: llamafile bundles model + runtime into one executable
+2. **OpenAI-compatible**: Drop-in replacement for OpenAI API at localhost:8080
+3. **GPU acceleration**: Auto-detects Metal (macOS), CUDA (Linux), Vulkan
+4. **Luna integration**: `--local` flag on any Luna command routes through llamafile
+5. **Model caching**: Downloaded models are cached in `~/.llamafile/models/`
+
+## Available Models
+
+| Model | Size | Speed | Best For |
+|-------|------|-------|----------|
+| qwen3.5-0.8b | 800 MB | Very fast | Quick tasks, offline dev |
+| **gemma4:e2b** | **3 GB** | **Very fast** | **Edge, audio+vision, 128K ctx** |
+| **gemma4:e4b** | **5 GB** | **Fast** | **Edge, audio+vision, 128K ctx** |
+| mistral-7b | 7 GB | Fast | General purpose |
+| llama-3.1-8b | 8 GB | Medium | Balanced quality |
+| codellama-13b | 13 GB | Slower | Code generation |
+| **gemma4:26b** | **15 GB** | **Fast (MoE)** | **Best open-source, 256K ctx, 3.8B active** |
+| **gemma4:31b** | **18 GB** | **Medium** | **Highest quality open model, 256K ctx** |
+
+## Usage
+
+```bash
+/local-llm start                                 # Start with default model
+/local-llm start --model llama-3.1-8b            # Start with specific model
+/local-llm status                                # Check if running
+/local-llm stop                                  # Shut down server
+```
+
+## Luna CLI Integration
+
+```bash
+luna run --local "Write unit tests for auth"     # Uses llamafile
+luna chain --local plan >> go >> test             # Full pipeline offline
+luna agent --local "Refactor this module"         # Agent uses local LLM
+```
+
+## Use Cases
+
+| Use Case | Why |
+|----------|-----|
+| Offline development | No internet required |
+| Free tier | Zero API costs |
+| Air-gapped environments | Security-sensitive work |
+| Privacy | Code never leaves your machine |
+| CI runners | No API key management |
+
+## Output Structure
+
+```
+~/.llamafile/
+├── bin/llamafile           # Binary
+├── models/                 # Cached model files
+│   ├── qwen3.5-0.8b.gguf
+│   └── llama-3.1-8b.gguf
+└── logs/                   # Server logs
+    └── llamafile.log
+```
+
+## In Pipes
+
+```bash
+/pipe local-llm start >> go *5 >> test >> local-llm stop     # Offline sprint
+/pipe local-llm start >> swarm "build feature" --agents 3    # Local swarm
+/pipe local-llm status >> if $running >> go >> else >> local-llm start >> go
+```
+
+## Reference
+
+- llamafile: https://github.com/mozilla-ai/llamafile
+- Models: https://huggingface.co/models?library=gguf

package/commands/ll-ls-products.md

@@ -0,0 +1,160 @@
+---
+name: ll-ls-products
+displayName: Luna LemonSqueezy Products
+description: Scan project and generate an HTML page with LemonSqueezy product definitions — descriptions, pricing tiers, features, and AI image prompts for hero, icon, social card, thumbnail, and variant cards
+version: 2.0.0
+category: monetization
+parameters:
+  - name: output_path
+    type: string
+    description: Where to save the HTML file
+    required: false
+    prompt: false
+workflow:
+  - analyze_project
+  - define_products_and_tiers
+  - generate_image_prompts
+  - write_html_file
+output:
+  - scripts/ls-product-descriptions.html (or custom path)
+prerequisites: []
+---
+
+# Luna LemonSqueezy Product Page Generator
+
+Scan the current project, define products with pricing tiers, generate AI image prompts, and **write a complete HTML file** where every field is click-to-copy — ready to paste into LemonSqueezy.
+
+## Step 1 — Analyze the Project
+
+Read these files in the current project:
+
+- `package.json` — name, description
+- `README.md` — features, tagline, value prop
+- `CLAUDE.md` — mission, target user, architecture
+- Marketing / landing pages — hero copy, feature lists, brand colors
+- Any existing pricing, billing config, or plan definitions
+- Source structure — to identify capabilities worth gating by tier
+
+Extract:
+- **Product name** and short tagline
+- **Selling description** (2-3 sentences)
+- **Target segments**: solo dev, pro user, team, enterprise
+- **Feature list** ranked by customer value
+- **Gating dimensions**: API calls, seats, instances, storage, retention, support SLA
+- **Brand accent color** (default violet #8b5cf6 if not found)
+
+## Step 2 — Define Products and Tiers
+
+For each product, create 2-4 pricing tiers:
+
+| Field | Content |
+|-------|---------|
+| Variant Name | `{Product} {Tier}` e.g. "MyApp Pro" |
+| Price | Monthly USD integer e.g. `29` |
+| Description | 2-3 sentences: what's included, limits, support |
+| Features | 6-10 checkmark bullets |
+
+Pricing rules:
+- 3 tiers is ideal, 2-4 acceptable
+- ~3-4x price jump between tiers
+- Middle tier marked "Most Popular"
+- Each tier unlocks meaningful new capabilities
+
+## Step 3 — Generate AI Image Prompts
+
+Generate **6 prompts total**: 4 product-level + 1 per variant tier.
+
+Every prompt uses the same layout: **left third = visual icon/mark, right two-thirds = product name + value prop + feature details** with clear typography hierarchy.
+
+### Product-Level Prompts (4)
+
+**Hero (1270x760)**
+Left third: product icon/symbol with accent glow.
+Right two-thirds: "[Product Name]" headline in large white bold, tagline in gray below, 3 small icon-label feature pairs in accent color, "Start Free" pill CTA at bottom right. Black bg (#0a0a0f). Apple-quality.
+
+**App Icon (1024x1024)**
+Upper half: product icon/symbol centered with accent glow.
+Lower half: "[Product Name]" in white bold, thin accent line separating icon from text.
+Black bg, rounded corners. Works at 32px. SF Symbol simplicity.
+
+**Open Graph (1200x630)**
+Left third: product icon/symbol with light beam fanning right, illuminating faint holographic UI outlines.
+Right two-thirds: "[Product Name]" headline, tagline in gray, feature labels in accent.
+Grid at 3% opacity. Cinematic, premium.
+
+**LS Thumbnail (800x400)**
+Left third: product icon/symbol with accent glow.
+Right two-thirds: "[Product Name]" headline, tagline, feature labels.
+Clean, Apple-inspired spacing.
+
+### Variant Tier Prompts (1 per tier, 800x400)
+
+Left third: tier-scaled icon (single for personal, with orbital ring for pro, 3 in triangle for team/enterprise). Accent glow.
+Right two-thirds: small accent label "[Tier]" (+ "Most Popular" badge if middle), large "$X/mo" headline in white bold, primary limit in gray, feature details in dimmer gray.
+
+## Step 4 — Write the HTML File
+
+Write a single self-contained HTML file to output path (default: `scripts/ls-product-descriptions.html`). Create dirs if needed.
+
+### Structure
+
+```
+body > .container
+  h1 "[Product] — LemonSqueezy Product Setup"
+  p.subtitle with link to app.lemonsqueezy.com/products/new
+  .instructions (how-to-use steps)
+  .product
+    .product-header (h2 name + badge)
+    .variant "Product Setup"
+      .field: Product Name (click-to-copy)
+      .field: Product Description (click-to-copy)
+      .ai-prompt-label + .ai-prompt: Hero Image (click-to-copy)
+      .ai-prompt-label + .ai-prompt: App Icon (click-to-copy)
+      .ai-prompt-label + .ai-prompt: Open Graph (click-to-copy)
+      .ai-prompt-label + .ai-prompt: LS Thumbnail (click-to-copy)
+    .variant (repeat per tier)
+      .field: Variant Name (click-to-copy)
+      .field: Price (click-to-copy) + "USD, Monthly" hint
+      .field: Variant Description (click-to-copy)
+      ul.features-list (checkmark items)
+      .ai-prompt-label + .ai-prompt: Variant card (click-to-copy)
+  script: copyThis()
+```
+
+### Styling
+
+```css
+body        — #0a0a0a bg, #e5e5e5 text, system font, 40px padding
+.container  — max-width 900px, centered
+.product    — #222 border, 12px radius
+.product-header — #141414 bg, #222 border-bottom
+.field-value    — #111 bg, #222 border, hover #444
+.field-value.copied — #00E5C3 border + "Copied!" top-right label
+.variant-price  — #00E5C3 teal
+.ai-prompt      — #0d1117 bg, #1a2332 border, monospace font
+.ai-prompt.copied — #00E5C3 border + "Copied!" label
+.ai-prompt-label — uppercase, small, #666 text, above prompt
+.features-list li::before — "✓ " in #2ECC7B
+```
+
+### JavaScript
+
+```javascript
+function copyThis(el) {
+  navigator.clipboard.writeText(el.innerText).then(() => {
+    el.classList.add('copied');
+    setTimeout(() => el.classList.remove('copied'), 1500);
+  });
+}
+```
+
+Every `.field-value` and `.ai-prompt` has `onclick="copyThis(this)"`.
+
+## After Generation
+
+Tell the user:
+1. Open the HTML file in a browser
+2. Go to [app.lemonsqueezy.com/products/new](https://app.lemonsqueezy.com/products/new)
+3. Click any field to copy, paste into the LS form
+4. Create product first with name + description, then add variant tiers
+5. Use the AI image prompts to generate product images in any AI image tool