specweave 0.6.7 → 0.7.0

package/src/adapters/gemini/adapter.ts

@@ -1,298 +0,0 @@
-/**
- * Gemini CLI Adapter
- *
- * Semi-automation adapter for Google Gemini CLI.
- * Uses universal AGENTS.md file for instructions.
- *
- * Gemini CLI features:
- * - Open-source AI agent (Apache 2.0)
- * - 1M token context window
- * - ReAct loop with built-in tools
- * - MCP (Model Context Protocol) support
- * - Agent mode for multi-file tasks
- */
-
-import * as path from 'path';
-import fs from 'fs-extra';
-import { AdapterBase } from '../adapter-base.js';
-import { AdapterOptions, AdapterFile } from '../adapter-interface.js';
-import { AgentsMdGenerator } from '../agents-md-generator.js';
-import { getDirname } from '../../utils/esm-helpers.js';
-
-const __dirname = getDirname(import.meta.url);
-
-export class GeminiAdapter extends AdapterBase {
-  name = 'gemini';
-  description = 'Gemini CLI adapter - Semi-automation with AGENTS.md and agent mode';
-  automationLevel = 'semi' as const;
-
-  /**
-   * Detect if Gemini CLI is available
-   *
-   * Checks for:
-   * - gemini command in PATH
-   * - .gemini/ directory exists
-   * - AGENTS.md file exists (previously configured)
-   */
-  async detect(): Promise<boolean> {
-    const hasGeminiCLI = await this.commandExists('gemini');
-    const hasGeminiDir = await this.fileExists('.gemini');
-    const hasAgentsMd = await this.fileExists('AGENTS.md');
-
-    return hasGeminiCLI || hasGeminiDir || hasAgentsMd;
-  }
-
-  /**
-   * Get files to install for Gemini adapter
-   */
-  getFiles(): AdapterFile[] {
-    return [
-      {
-        sourcePath: 'AGENTS.md',
-        targetPath: 'AGENTS.md',
-        description: 'Universal SpecWeave instructions (works with all AI tools)'
-      },
-      {
-        sourcePath: 'README.md',
-        targetPath: '.gemini/README.md',
-        description: 'Gemini adapter documentation'
-      }
-    ];
-  }
-
-  /**
-   * Install Gemini adapter
-   */
-  async install(options: AdapterOptions): Promise<void> {
-    console.log('\n📦 Installing Gemini CLI Adapter (Semi-Automation)\n');
-
-    // Ensure .gemini directory exists
-    const geminiDir = path.join(options.projectPath, '.gemini');
-    await fs.ensureDir(geminiDir);
-
-    // Generate AGENTS.md
-    const agentsMdPath = path.join(options.projectPath, 'AGENTS.md');
-    await this.generateAgentsMd(agentsMdPath, options);
-
-    // Copy README
-    const readmePath = path.join(__dirname, 'README.md');
-    if (await fs.pathExists(readmePath)) {
-      await fs.copy(readmePath, path.join(geminiDir, 'README.md'));
-    }
-
-    console.log('\n✨ Gemini adapter installed!');
-    console.log('\n📋 Files created:');
-    console.log('   - AGENTS.md (universal instructions)');
-    console.log('   - .gemini/README.md (adapter documentation)');
-  }
-
-  /**
-   * Generate AGENTS.md file
-   */
-  private async generateAgentsMd(targetPath: string, options: AdapterOptions): Promise<void> {
-    const generator = new AgentsMdGenerator(
-      path.join(__dirname, '../../skills'),
-      path.join(__dirname, '../../agents'),
-      path.join(__dirname, '../../commands')
-    );
-
-    const content = await generator.generate({
-      projectName: options.projectName,
-      projectPath: options.projectPath
-    });
-
-    await fs.writeFile(targetPath, content);
-    console.log('  ✅ AGENTS.md - Universal SpecWeave instructions');
-  }
-
-  /**
-   * Post-installation instructions
-   */
-  async postInstall(options: AdapterOptions): Promise<void> {
-    console.log(this.getInstructions());
-  }
-
-  /**
-   * Get usage instructions for Gemini adapter
-   */
-  getInstructions(): string {
-    return `
-================================================================
-      Gemini CLI Adapter - Semi-Automation
-================================================================
-
-Your project is now configured for Gemini CLI with SEMI-automation!
-
-WHAT THIS PROVIDES:
-
-- AGENTS.md (Universal Instructions)
-  - Works with Gemini CLI AND other tools
-  - Auto-generated from actual skills/agents
-  - SpecWeave structure and workflows
-  - 1M token context window support
-
-- Agent Mode Integration
-  - Multi-file task handling
-  - ReAct loop with built-in tools
-  - MCP support for extensibility
-
-QUICK START:
-
-1. Install Gemini CLI (if not already installed):
-
-   npm install -g @google-cloud/gemini-cli
-
-2. Authenticate:
-
-   gemini auth
-
-3. Create your first feature:
-
-   gemini "Read AGENTS.md and create increment for user authentication"
-
-   Gemini will:
-   - Read AGENTS.md for SpecWeave context
-   - Adopt PM role: Create spec.md (WHAT/WHY)
-   - Adopt Architect role: Create plan.md (HOW)
-   - Create tasks.md
-   - Use agent mode for multi-file operations
-
-4. Continue development:
-
-   gemini "Implement task T001 from increment 0001"
-   gemini "Run tests for increment 0001"
-   gemini "Review and fix any issues"
-
-HOW TO USE:
-
-Gemini CLI reads AGENTS.md automatically when you reference it:
-
-\`\`\`bash
-# Explicit reference (recommended):
-gemini "Read AGENTS.md and [your request]"
-
-# Natural language:
-gemini "Create a new feature following SpecWeave structure"
-
-# With context:
-gemini "Following the PM role in AGENTS.md, create requirements for payments"
-\`\`\`
-
-CONTEXT LOADING (70%+ Token Savings):
-
-CRITICAL: Tell Gemini to read context-manifest.yaml first!
-
-\`\`\`bash
-gemini "Read context-manifest.yaml from increment 0001, load only those files, then implement task T001"
-\`\`\`
-
-**Why this matters**:
-- Gemini has 1M token context window
-- But loading everything wastes tokens
-- Context manifests = precise, efficient loading
-
-FEATURES:
-
-✅ 1M Token Context Window
-   - Can handle large specifications
-   - Read entire codebase if needed
-
-✅ Agent Mode
-   - Multi-file task completion
-   - Automatic planning and execution
-   - Interactive approval before changes
-
-✅ Built-in Tools
-   - File operations
-   - Shell commands
-   - Google Search grounding
-   - Web fetching
-
-✅ MCP Support
-   - Extend with custom integrations
-   - Local or remote MCP servers
-
-✅ Free Tier
-   - Gemini 2.5 Pro included
-   - 60 requests/minute
-   - 1,000 requests/day
-
-EXAMPLE WORKFLOWS:
-
-1. **Create Feature**:
-   \`\`\`bash
-   gemini "Read AGENTS.md. Create increment 0002 for payment processing with Stripe. Follow SpecWeave structure: spec.md (PM role), plan.md (Architect role), tasks.md"
-   \`\`\`
-
-2. **Implement Task**:
-   \`\`\`bash
-   gemini "Read increment 0002, implement task T001 (Setup Stripe SDK)"
-   \`\`\`
-
-3. **Fix Bug**:
-   \`\`\`bash
-   gemini "Read AGENTS.md and increment 0001. Fix authentication bug in login endpoint. Update tests."
-   \`\`\`
-
-4. **Review Code**:
-   \`\`\`bash
-   gemini "Adopt Tech Lead role from AGENTS.md. Review code in src/auth/ and provide feedback"
-   \`\`\`
-
-TIPS & TRICKS:
-
-1. **Always Reference AGENTS.md**
-   - Ensures Gemini understands SpecWeave structure
-   - Provides role definitions (PM, Architect, etc.)
-
-2. **Use Agent Mode**
-   - For complex, multi-file tasks
-   - Gemini proposes plan before execution
-   - You approve changes
-
-3. **Leverage 1M Context Window**
-   - Can load large specs
-   - But still use context manifests for efficiency
-
-4. **MCP Integration**
-   - Connect to your tools (JIRA, GitHub, etc.)
-   - See: https://modelcontextprotocol.io/
-
-5. **Check Gemini Documentation**
-   - \`gemini --help\`
-   - https://developers.google.com/gemini-code-assist/docs/gemini-cli
-
-COMPARISON:
-
-| Feature | Claude Code | Gemini CLI |
-|---------|-------------|------------|
-| **Automation** | Full | Semi |
-| **Skills** | Native | Via AGENTS.md |
-| **Agents** | Native | Via AGENTS.md |
-| **Context Window** | 200k | 1M (5x larger!) |
-| **Cost** | Paid | Free tier available |
-| **Open Source** | No | Yes (Apache 2.0) |
-
-LIMITATIONS:
-
-- No native skills/agents (manual role adoption via AGENTS.md)
-- No hooks (can't auto-update docs on events)
-- Requires explicit instruction referencing
-- CLI-only (no IDE integration yet)
-
-But still VERY powerful with 1M context and agent mode!
-
-DOCUMENTATION:
-
-- AGENTS.md: Universal SpecWeave instructions (READ THIS!)
-- SPECWEAVE.md: Complete framework documentation
-- .gemini/README.md: Gemini adapter guide
-- https://developers.google.com/gemini-code-assist/docs/gemini-cli
-
-You're ready to build with SpecWeave on Gemini CLI!
-
-Pro tip: Leverage the 1M token context window for large specifications,
-but always use context manifests for efficiency!
-    `;
-  }
-}

package/src/adapters/generic/README.md

@@ -1,277 +0,0 @@
-# Generic Adapter
-
-**Automation Level**: Manual (Works with literally ANY AI tool)
-
-## Overview
-
-The Generic adapter provides a **manual workflow** for SpecWeave that works with **100% of AI tools** - ChatGPT web, Claude web, Gemini, Perplexity, You.com, or any AI that can read text and follow instructions.
-
-## Key Value: Universal Compatibility
-
-**The Problem**:
-- Claude Code adapter: Only works with Claude Code (~10% market)
-- Cursor adapter: Only works with Cursor (~30% market)
-- Copilot adapter: Only works with VS Code + Copilot (~40% market)
-- **Combined: Still missing ~20% of developers!**
-
-**The Solution**:
-- Generic adapter: Works with **EVERYTHING** (100% market coverage)
-- ChatGPT web? ✅
-- Claude web? ✅
-- Gemini? ✅
-- Any other AI? ✅
-
-## What This Adapter Provides
-
-### SPECWEAVE-MANUAL.md (Complete Step-by-Step Guide)
-- **Step 1**: Create increment folder (terminal command)
-- **Step 2**: Create spec.md (copy template → paste to AI → save response)
-- **Step 3**: Create plan.md (copy template → paste to AI → save response)
-- **Step 4**: Create tasks.md (copy template → paste to AI → save response)
-- **Step 5**: Create context-manifest.yaml (fill manually)
-
-**Total time**: 30-60 minutes per feature (vs 30 seconds with Claude Code)
-
-### Trade-Off: Speed vs Compatibility
-
-| Adapter | Speed | Compatibility | Automation |
-|---------|-------|---------------|------------|
-| **Claude Code** | ⚡⚡⚡⚡⚡ (30s) | Claude only (~10%) | Full |
-| **Cursor** | ⚡⚡⚡⚡ (5min) | Cursor only (~30%) | Semi |
-| **Copilot** | ⚡⚡⚡ (15min) | VS Code+Copilot (~40%) | Basic |
-| **Generic** | ⚡ (30-60min) | **100% (ANY AI!)** | Manual |
-
-**Generic = Slowest, but works with EVERYTHING!**
-
-## Understanding "Manual" Workflow
-
-**Manual ≠ Hard, Manual = YOU Orchestrate Each Step**
-
-**Example workflow**:
-1. **You**: Open SPECWEAVE-MANUAL.md, read Step 2 (Create spec.md)
-2. **You**: Copy template prompt from manual
-3. **You**: Paste to ChatGPT (or any AI)
-4. **AI**: Generates spec.md content following SpecWeave patterns
-5. **You**: Copy AI's response
-6. **You**: Save to `spec.md` file manually
-7. **Repeat for plan.md, tasks.md, etc.**
-
-**Result**: Same quality as automated adapters, just more manual steps!
-
-## Installation
-
-```bash
-# Install SpecWeave with Generic adapter
-npx specweave init my-project --adapter generic
-
-# Files created:
-# SPECWEAVE-MANUAL.md    (complete step-by-step guide)
-```
-
-## Directory Structure
-
-```
-SPECWEAVE-MANUAL.md    # Main manual workflow guide (read this!)
-.specweave/
-└── adapters/
-    └── generic/
-        └── README.md   # This file
-```
-
-## Usage Example (Creating Authentication Feature)
-
-### Complete Workflow (30-60 minutes)
-
-**Step 1** (Terminal - 1 minute):
-```bash
-mkdir -p .specweave/increments/0001-user-authentication
-cd .specweave/increments/0001-user-authentication
-```
-
-**Step 2** (Create spec.md - 10 minutes):
-1. Open SPECWEAVE-MANUAL.md
-2. Find "Step 2: Create spec.md"
-3. Copy entire prompt template
-4. Paste to ChatGPT/Claude/Gemini
-5. Fill in your requirements: "Email/password auth, Google OAuth, JWT tokens..."
-6. AI generates complete spec.md
-7. Copy response → Save to `spec.md`
-
-**Step 3** (Create plan.md - 15 minutes):
-1. Copy plan.md template from manual
-2. Paste to AI along with spec.md content
-3. AI generates technical plan
-4. Copy response → Save to `plan.md`
-
-**Step 4** (Create tasks.md - 10 minutes):
-1. Copy tasks.md template from manual
-2. Paste to AI along with plan.md content
-3. AI generates implementation checklist
-4. Copy response → Save to `tasks.md`
-
-**Step 5** (Create context-manifest.yaml - 5 minutes):
-1. Manually create file
-2. List only relevant docs (auth specs, auth architecture)
-3. Save
-
-**Total**: 41 minutes (vs 30 seconds with Claude Code automation!)
-
-## Simulating Skills Manually
-
-### What are Skills?
-
-**In Claude Code**: Auto-activating capabilities
-**In Generic**: YOU follow skill workflows manually
-
-### Example: increment-planner Skill
-
-**Claude Code (automatic)**:
-```
-User: "create increment for auth"
-→ increment-planner skill auto-activates
-→ Creates spec, plan, tasks in 30 seconds
-```
-
-**Generic (manual)**:
-```
-You: Read SPECWEAVE-MANUAL.md
-You: Follow Step 1 (create folder - 1 min)
-You: Follow Step 2 (create spec.md with AI - 10 min)
-You: Follow Step 3 (create plan.md with AI - 15 min)
-You: Follow Step 4 (create tasks.md with AI - 10 min)
-You: Follow Step 5 (create manifest - 5 min)
-Total: 41 minutes
-```
-
-You manually execute the entire workflow!
-
-## Simulating Agents Manually
-
-### What are Agents?
-
-**In Claude Code**: Specialized roles with separate context windows
-**In Generic**: YOU tell AI which role to adopt
-
-### Example: PM Agent
-
-**Claude Code (automatic)**:
-```typescript
-Task({ subagent_type: "pm", prompt: "create spec" })
-→ PM agent creates spec.md
-```
-
-**Generic (manual)**:
-```
-You paste to ChatGPT:
-"Act as Product Manager and create spec.md.
-
-As PM, focus on:
-- WHAT and WHY (not HOW)
-- Technology-agnostic requirements
-- User stories with acceptance criteria
-
-Requirements: [your requirements]"
-```
-
-The AI adopts PM perspective and creates spec.md!
-
-### All Roles You Can Simulate
-
-| Role | Prompt Template | Focus |
-|------|----------------|-------|
-| **PM** | "Act as PM and create spec" | WHAT/WHY, user stories |
-| **Architect** | "Act as Architect and create plan" | HOW, technical design |
-| **DevOps** | "Act as DevOps and create infrastructure" | Deployment, monitoring |
-| **QA Lead** | "Act as QA and create test strategy" | Test coverage, cases |
-| **Security** | "Act as Security and review" | Threats, vulnerabilities |
-
-## Comparison with Other Adapters
-
-| Feature | Claude Code | Cursor | Copilot | Generic |
-|---------|-------------|--------|---------|---------|
-| **Automation** | Full | Semi | Basic | Manual |
-| **Skills** | Native | Simulated | N/A | Manual (follow guide) |
-| **Agents** | Native | Manual roles | N/A | Manual (tell AI role) |
-| **Context** | Auto-load | @ shortcuts | Instructions | Manual (copy files) |
-| **Speed** | ⚡⚡⚡⚡⚡ | ⚡⚡⚡⚡ | ⚡⚡⚡ | ⚡ |
-| **Compatibility** | 10% | 30% | 40% | **100%** |
-
-**Generic = Slowest but most compatible!**
-
-## When to Use This Adapter
-
-✅ **Use Generic adapter if**:
-- You use ChatGPT web, Claude web, Gemini, or other non-IDE AI
-- You don't have Claude Code, Cursor, or VS Code
-- You want maximum tool flexibility
-- Simple projects (1-5 increments)
-- Learning SpecWeave methodology
-
-⚠️ **Consider alternatives if**:
-- You have Claude Code → Use Claude adapter (10-20x faster)
-- You have Cursor → Use Cursor adapter (5-10x faster)
-- You have VS Code+Copilot → Use Copilot adapter (2-3x faster)
-- Large projects (10+ increments) → Manual becomes tedious
-
-## Context Manifests (70%+ Token Savings)
-
-**Same benefit as automated adapters!**
-
-### The Problem
-- Your specs might be 500+ pages
-- Loading all = 50k tokens
-- ChatGPT/Claude have token limits
-
-### The Solution
-- context-manifest.yaml lists ONLY relevant files
-- Load only 50 pages = 5k tokens
-- **Savings: 90%!**
-
-### How to Use with Your AI
-
-When implementing auth feature:
-1. Open `.specweave/increments/0001-auth/context-manifest.yaml`
-2. See relevant files listed:
-   ```yaml
-   spec_sections:
-     - .specweave/docs/internal/strategy/auth/spec.md
-   documentation:
-     - .specweave/docs/internal/architecture/auth/design.md
-   ```
-3. ONLY copy those 2 files to AI (not entire docs/ folder!)
-4. AI has relevant context without token waste
-
-**Result**: Same 70%+ token savings as automated adapters!
-
-## Tips for Success
-
-### 1. Follow SPECWEAVE-MANUAL.md Exactly
-Templates are tested and proven → Use as-is
-
-### 2. Use Context Manifests Always
-Don't paste 500 pages → Use manifests (list 50 pages only)
-
-### 3. Tell AI Which Role to Adopt
-"Act as PM" or "Act as Architect" → Better results
-
-### 4. Save AI Responses Immediately
-Don't lose generated content → Copy and save right away
-
-### 5. Iterate if Needed
-First pass not perfect? Ask AI to refine/improve
-
-## Related Documentation
-
-- [SPECWEAVE-MANUAL.md](SPECWEAVE-MANUAL.md) - **Read this first!** Complete step-by-step guide
-- [SPECWEAVE.md](../../SPECWEAVE.md) - Complete technical documentation
-- [Adapter Architecture](../README.md) - Multi-tool design philosophy
-
----
-
-**Status**: Active (v0.2.0-beta.1+)
-**Market Share**: ~20% (users without IDE-integrated AI)
-**Priority**: P1 (ensures 100% market coverage)
-
----
-
-**Remember**: Manual = Slower, but works with ANY AI tool! 🌍