@su-record/vibe 2.5.3 → 2.5.5

package/commands/vibe.spec.md

@@ -106,49 +106,23 @@ Collect requirements through conversation with the user and create an **AI-execu
 
 > **PTCF**: Persona, Task, Context, Format - Google Gemini prompt optimization framework
 
-## External LLM Integration (Optional)
+## External LLM Integration (GPT/Gemini)
 
-When external LLMs are enabled, automatically utilize during SPEC creation:
+When GPT/Gemini are enabled, they are automatically utilized during SPEC creation:
 
 ```
-/vibe.spec "complex feature"
+/vibe.spec "feature"
       ↓
-[Claude Opus] Create SPEC draft
+[Claude] Draft SPEC
       ↓
-[GPT enabled?] → Call global hook script via Bash
+[Parallel Research] GPT + Gemini + Claude agents (8 parallel)
       ↓
-[Gemini enabled?] → Call global hook script via Bash
+[SPEC Review] GPT + Gemini parallel review
       ↓
 [Claude] Finalize SPEC
 ```
 
-| External LLM | Invocation Method | Role | When Used |
-|--------------|-------------------|------|-----------|
-| GPT (user query) | `gpt-`, `gpt.`, `지피티-` prefix | Direct question (Web Search enabled) | User asks directly |
-| GPT (orchestration) | Call global hook script via Bash | Internal orchestration (JSON, no search) | SPEC/vibe.run internal |
-| Gemini (user query) | `gemini-`, `gemini.`, `제미나이-` prefix | Direct question (Google Search enabled) | User asks directly |
-| Gemini (orchestration) | Call global hook script via Bash | Internal orchestration (JSON, no search) | SPEC/vibe.run internal |
-
-**Claude internal orchestration call (Bash):**
-```bash
-# Usage: node llm-orchestrate.js <provider> <mode> [systemPrompt] [prompt]
-#   - If systemPrompt is omitted, default value is used
-#   - If "-" is passed for systemPrompt, default value is used and next argument is treated as prompt
-
-# Cross-platform path resolution (recommended)
-# Windows uses %APPDATA%, bash uses $APPDATA - Node.js handles both via process.env.APPDATA
-
-# GPT call
-node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gpt orchestrate-json "[question content]"
-
-# Gemini call
-node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gemini orchestrate-json "[question content]"
-
-# Custom system prompt usage
-node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gpt orchestrate-json "You are a SPEC reviewer" "[question content]"
-```
-
-**Activation:**
+**Setup:**
 ```bash
 vibe gpt auth       # Enable GPT (OAuth)
 vibe gemini auth    # Enable Gemini (OAuth)
@@ -264,7 +238,14 @@ Read ~/.claude/vibe/languages/typescript-react.md
 
 ### 3. Parallel Research (v2.5.0) - MANDATORY AFTER requirements confirmed
 
-**🚨 CRITICAL: Research is MANDATORY after requirements are confirmed**
+**🚨🚨🚨 ABSOLUTE RULE: YOU MUST USE BASH TOOL FOR RESEARCH 🚨🚨🚨**
+
+**STOP! Before doing ANY research, read this carefully:**
+
+1. **DO NOT** use Task tool to spawn research agents
+2. **DO NOT** use context7 MCP directly for research
+3. **DO NOT** use WebSearch tool directly for research
+4. **YOU MUST** use Bash tool to call llm-orchestrate.js directly
 
 **When to trigger:**
 1. ✅ Feature type decided (e.g., "passkey authentication")
@@ -272,87 +253,54 @@ Read ~/.claude/vibe/languages/typescript-react.md
 3. ✅ Language guide copied (step 2.5)
 4. ✅ Core requirements collected
 
-**→ IMMEDIATELY run orchestrator research via Bash. NO EXCEPTIONS.**
-
-**🚨 MANDATORY: Use Bash tool to call orchestrator**
-
-❌ **FORBIDDEN:** Using Task tool to spawn agents directly
-✅ **REQUIRED:** Use Bash tool to run the orchestrator command below
-
-**Execution via Bash (REQUIRED):**
-```bash
-node -e "import('@su-record/vibe/orchestrator').then(o => o.research('[FEATURE]', ['[STACK1]', '[STACK2]']).then(r => console.log(r.content[0].text)))"
-```
+**→ IMMEDIATELY run these 4 Bash commands IN PARALLEL (all at once):**
 
-**Example (copy and run via Bash tool):**
 ```bash
-# After confirming: passkey auth + React + Supabase
-node -e "import('@su-record/vibe/orchestrator').then(o => o.research('passkey authentication', ['React', 'Supabase']).then(r => console.log(r.content[0].text)))"
-```
+# Cross-platform path (works on Windows/macOS/Linux)
+VIBE_SCRIPTS="$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts"
 
-**What runs in parallel (via orchestrator):**
+# 1. GPT: Best practices
+node "$VIBE_SCRIPTS/llm-orchestrate.js" gpt orchestrate-json "Best practices for [FEATURE] with [STACK]. Focus: architecture patterns, code conventions. Return JSON: {patterns: [], antiPatterns: [], libraries: []}"
 
-| Component | Role | Source |
-|-----------|------|--------|
-| Claude Agent: `best-practices-agent` | Best practices for [feature] + [stack] | WebSearch |
-| Claude Agent: `framework-docs-agent` | Latest docs via context7 | context7 MCP |
-| Claude Agent: `codebase-patterns-agent` | Similar patterns in existing codebase | Glob, Grep |
-| Claude Agent: `security-advisory-agent` | Security advisories for [feature] | WebSearch |
-| **GPT: Best Practices** | Architecture patterns, code conventions | GPT API |
-| **GPT: Security** | CVE database, vulnerability details | GPT API |
-| **Gemini: Best Practices** | Latest trends, framework updates | Gemini API |
-| **Gemini: Security** | Latest security advisories, patches | Gemini API |
+# 2. GPT: Security
+node "$VIBE_SCRIPTS/llm-orchestrate.js" gpt orchestrate-json "Security vulnerabilities for [FEATURE] with [STACK]. Focus: CVE database, known exploits. Return JSON: {vulnerabilities: [], mitigations: [], checklist: []}"
 
-#### 3.1 Multi-LLM Research Enhancement (v2.5.0)
+# 3. Gemini: Best practices
+node "$VIBE_SCRIPTS/llm-orchestrate.js" gemini orchestrate-json "Best practices for [FEATURE] with [STACK]. Focus: latest trends, framework updates. Return JSON: {patterns: [], antiPatterns: [], libraries: []}"
 
-**vibe = Quality Assurance Framework**
+# 4. Gemini: Security
+node "$VIBE_SCRIPTS/llm-orchestrate.js" gemini orchestrate-json "Security advisories for [FEATURE] with [STACK]. Focus: latest patches, recent incidents. Return JSON: {advisories: [], patches: [], incidents: []}"
+```
 
-When GPT/Gemini are enabled, research agents use **3 LLM perspectives** for critical areas:
+**Concrete example - run all 4 in parallel:**
+```bash
+# GPT best practices
+node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gpt orchestrate-json "Best practices for passkey authentication with React, Supabase. Focus: architecture patterns, code conventions. Return JSON: {patterns: [], antiPatterns: [], libraries: []}"
 
-```
-┌─────────────────────────────────────────────────────────────┐
-│  MULTI-LLM PARALLEL RESEARCH                                │
-├─────────────────────────────────────────────────────────────┤
-│  best-practices-agent                                       │
-│    ├── Claude (Haiku): Core patterns, anti-patterns         │
-│    ├── GPT: Architecture patterns, code conventions         │
-│    └── Gemini: Latest trends, framework updates             │
-├─────────────────────────────────────────────────────────────┤
-│  security-advisory-agent                                    │
-│    ├── Claude (Haiku): OWASP Top 10, security patterns      │
-│    ├── GPT: CVE database, vulnerability details             │
-│    └── Gemini: Latest security advisories, patches          │
-└─────────────────────────────────────────────────────────────┘
-        ↓
-    Merge & Dedupe (per agent)
-        ↓
-    SPEC Context + Constraints
-```
+# GPT security
+node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gpt orchestrate-json "Security vulnerabilities for passkey authentication with React, Supabase. Focus: CVE database, known exploits. Return JSON: {vulnerabilities: [], mitigations: [], checklist: []}"
 
-**Parallel execution (run alongside Claude agents):**
+# Gemini best practices
+node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gemini orchestrate-json "Best practices for passkey authentication with React, Supabase. Focus: latest trends, framework updates. Return JSON: {patterns: [], antiPatterns: [], libraries: []}"
 
-```bash
-# Cross-platform path
-VIBE_SCRIPTS="$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts"
+# Gemini security
+node "$(node -p "process.env.APPDATA || require('os').homedir() + '/.config'")/vibe/hooks/scripts/llm-orchestrate.js" gemini orchestrate-json "Security advisories for passkey authentication with React, Supabase. Focus: latest patches, recent incidents. Return JSON: {advisories: [], patches: [], incidents: []}"
+```
 
-# Best practices - GPT perspective
-node "$VIBE_SCRIPTS/llm-orchestrate.js" gpt orchestrate-json \
-  "Best practices for [FEATURE] with [STACK]. Focus: architecture patterns, code conventions. Return JSON: {patterns: [], antiPatterns: [], libraries: []}"
+**ALSO run Claude research agents in parallel using Task tool:**
 
-# Best practices - Gemini perspective
-node "$VIBE_SCRIPTS/llm-orchestrate.js" gemini orchestrate-json \
-  "Best practices for [FEATURE] with [STACK]. Focus: latest trends, framework updates. Return JSON: {patterns: [], antiPatterns: [], libraries: []}"
+| Claude Agent | Role | Source |
+|--------------|------|--------|
+| `best-practices-agent` | Best practices for [feature] + [stack] | WebSearch |
+| `framework-docs-agent` | Latest docs via context7 | context7 MCP |
+| `codebase-patterns-agent` | Similar patterns in existing codebase | Glob, Grep |
+| `security-advisory-agent` | Security advisories for [feature] | WebSearch |
 
-# Security - GPT perspective
-node "$VIBE_SCRIPTS/llm-orchestrate.js" gpt orchestrate-json \
-  "Security vulnerabilities for [FEATURE] with [STACK]. Focus: CVE database, known exploits. Return JSON: {vulnerabilities: [], mitigations: [], checklist: []}"
+**Total: 4 GPT/Gemini calls (Bash) + 4 Claude agents (Task) = 8 parallel research tasks**
 
-# Security - Gemini perspective
-node "$VIBE_SCRIPTS/llm-orchestrate.js" gemini orchestrate-json \
-  "Security advisories for [FEATURE] with [STACK]. Focus: latest patches, recent incidents. Return JSON: {advisories: [], patches: [], incidents: []}"
-```
+**🚨 GPT/Gemini MUST be called via Bash with llm-orchestrate.js! 🚨**
 
-**Result merge rules:**
+#### 3.1 Result Merge Rules
 
 | Area | Merge Strategy |
 |------|----------------|
@@ -366,7 +314,7 @@ node "$VIBE_SCRIPTS/llm-orchestrate.js" gemini orchestrate-json \
 - ✅ ALWAYS run after requirements confirmed
 - ✅ Show "Running parallel research (Claude + GPT + Gemini)..." message
 - ✅ Include all agent + LLM results in SPEC Context
-- ✅ Run all LLM calls in parallel (4 Bash calls simultaneously)
+- ✅ Run all 4 Bash LLM calls in parallel + 4 Task agents in parallel
 
 **Research results are reflected in SPEC's Context section.**
 

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@su-record/vibe",
-  "version": "2.5.3",
+  "version": "2.5.5",
   "description": "Vibe - Claude Code exclusive SPEC-driven AI coding framework with 35+ integrated tools",
   "type": "module",
   "main": "dist/cli/index.js",

package/skills/vibe-capabilities.md

@@ -52,14 +52,6 @@ node -e "import('@su-record/vibe/tools').then(t => t.TOOL_NAME({...}))"
 
 ## Orchestrator
 
-### Parallel Research (4 agents)
-
-```bash
-node -e "import('@su-record/vibe/orchestrator').then(o =>
-  o.research('feature', ['stack1', 'stack2'])
-)"
-```
-
 ### Background Agents
 
 ```bash
@@ -68,15 +60,37 @@ node -e "import('@su-record/vibe/orchestrator').then(o =>
 )"
 ```
 
-## LLM Integrations (Hooks)
+## Multi-LLM Research (v2.5.0)
+
+GPT/Gemini are automatically called within `/vibe.spec`:
+
+```text
+/vibe.spec "feature"
+      ↓
+[Claude] Draft SPEC
+      ↓
+[Parallel Research] 8 parallel tasks:
+  - 4x Bash: GPT/Gemini (best practices + security)
+  - 4x Task: Claude agents (docs, patterns, advisories)
+      ↓
+[SPEC Review] GPT + Gemini parallel review
+      ↓
+[Claude] Finalize SPEC
+```
 
-Use prefix patterns to call GPT/Gemini:
+| Phase | Method | Tasks |
+|-------|--------|-------- |
+| Research | `llm-orchestrate.js` via Bash | GPT best practices, GPT security, Gemini best practices, Gemini security |
+| Research | Task tool (Claude agents) | framework-docs, codebase-patterns, best-practices, security-advisory |
+| SPEC Review | `llm-orchestrate.js` via Bash | SPEC quality validation |
+| Docs | context7 MCP | Latest library documentation |
 
-| LLM | Prefix | Example | Features |
-|-----|--------|---------|----------|
-| GPT | `gpt-`, `gpt.` | `gpt.today's weather in Seoul` | Web Search enabled |
-| Gemini | `gemini-`, `gemini.` | `gemini.Analyze UX` | Google Search enabled |
-| context7 | Plugin call | Library documentation | `/plugin install context7` |
+**Setup:**
+```bash
+vibe gpt auth       # Enable GPT
+vibe gemini auth    # Enable Gemini
+vibe status         # Check current settings
+```
 
 ## ULTRAWORK Mode
 
@@ -102,15 +116,22 @@ Add `ultrawork` or `ulw` for maximum performance:
 ## Project Structure
 
 ```
-.claude/
+# Global (~/.claude/)
+~/.claude/
 ├── commands/          # Slash commands
 ├── agents/            # Sub-agents
 ├── skills/            # Auto-activated guides
 └── vibe/
-    ├── specs/         # SPEC documents
-    ├── features/      # BDD scenarios
-    ├── rules/         # Coding rules
-    └── solutions/     # Documented solutions
+    ├── rules/         # Coding rules (global)
+    ├── languages/     # Language guides (global)
+    └── templates/     # Templates (global)
+
+# Project (.claude/vibe/)
+.claude/vibe/
+├── specs/             # SPEC documents
+├── features/          # BDD scenarios
+├── config.json        # Project config
+└── constitution.md    # Project rules
 ```
 
 ## Context Management
@@ -124,6 +145,5 @@ Add `ultrawork` or `ulw` for maximum performance:
 Even without `/vibe.*` commands, you can:
 
 1. Call tools directly via node commands
-2. Use LLM hooks (`gpt-`, `gemini-`)
-3. Reference skills for guidance
-4. Apply coding rules from `~/.claude/vibe/rules/` (global)
+2. Reference skills for guidance
+3. Apply coding rules from `~/.claude/vibe/rules/` (global)

package/skills/multi-llm-orchestration.md

@@ -1,90 +0,0 @@
----
-description: Multi-LLM collaboration guide. Auto-activates for architecture review, design decisions, debugging, UI/UX consultation, complex problem solving, or when a different perspective is needed.
----
-# Multi-LLM Orchestration
-
-Guide for using GPT/Gemini as sub-agents in Claude Code via hooks.
-
-## Calling GPT
-
-Use `gpt-` or `gpt.` prefix:
-
-| Situation | Example |
-|-----------|---------|
-| Architecture review | `gpt- Review this auth architecture: [code]` |
-| Complex debugging | `gpt- Debug this error: [error message]` |
-| Algorithm optimization | `gpt- Suggest better algorithm for: [problem]` |
-| Security review | `gpt- Check security vulnerabilities in: [code]` |
-| Technology decisions | `gpt- Compare Redis vs Memcached for caching` |
-
-### Usage Examples
-
-```
-gpt- Review this authentication architecture: [code]
-gpt.Review this REST API design
-```
-
-## Calling Gemini
-
-Use `gemini-` or `gemini.` prefix:
-
-| Situation | Example |
-|-----------|---------|
-| UI/UX consultation | `gemini- Improve UX for this login form` |
-| Design system advice | `gemini- Suggest component structure` |
-| Accessibility review | `gemini- Check a11y issues in this form` |
-| User flow design | `gemini- Optimize this checkout flow` |
-
-### Usage Examples
-
-```
-gemini- Improve UX for this login form
-gemini.Suggest improvements for this dashboard
-```
-
-## When NOT to Use
-
-- Simple implementation (Claude alone is sufficient)
-- File read/write (use Claude's tools)
-- Quick response needed (API latency exists)
-- Tasks requiring project context (Claude has the context)
-
-## Collaboration Patterns
-
-### Pattern 1: Design Cross-check
-
-```
-Claude: Draft design
-    ↓
-gpt- Review this architecture: [design]
-    ↓
-Claude: Incorporate feedback into final design
-```
-
-### Pattern 2: Debugging Collaboration
-
-```
-Claude: Attempt problem analysis
-    ↓
-If stuck: gpt- Debug this issue: [error details]
-    ↓
-Claude: Apply GPT's suggestions
-```
-
-### Pattern 3: UI/UX Consultation
-
-```
-Claude: Implement feature
-    ↓
-gemini- Suggest UX improvements for: [component]
-    ↓
-Claude: Apply UX feedback
-```
-
-## API Key Setup
-
-```bash
-vibe gpt login      # Enable GPT (OAuth)
-vibe gemini login   # Enable Gemini (OAuth)
-vibe status         # Check current settings
-```