@leejungkiin/awkit 1.0.9 → 1.1.1

package/skills/awf-session-restore/SKILL.md

@@ -1,14 +1,25 @@
 ---
 name: awf-session-restore
-description: Auto-restore context from Beads + Brain + Project Brain on session start
+description: |
+  Silent context restoration via Symphony + NeuralMemory with strict project scoping.
+  Gathers Git, Symphony, and Brain context silently — no console spam.
+  Enforces Project ID → Brain Switch → Memory Read order to prevent cross-project contamination.
+metadata:
+  stage: core
+  version: "7.0"
+  replaces: "v6.4"
+  requires: symphony-orchestrator
+  tags: [session, restore, context, symphony, neuralmemory, silent, multi-project]
 trigger: session_start
+invocation-type: auto
 priority: 2
 ---
 
-# AWF Session Restore (v6.4 - Project-Aware)
+# AWF Session Restore (v7.0 — Symphony Native)
 
-> **Purpose:** Tự động khôi phục context khi bắt đầu session.
-> **⛔ MANDATORY:** Luôn THỰC SỰ chạy `bd list` qua terminal — không chỉ đề cập.
+> **Purpose:** Silently gather unstructured context (Git, Plans, Memory) at session start.
+> **Key Change v7.0:** Symphony-native. Strict brain scoping.
+> **Output:** NO console block. Context injected silently for AI consumption.
 
 ---
 
@@ -21,237 +32,148 @@ Skill này **BẮT BUỘC** chạy khi:
 
 ---
 
-## Execution Flow
+## Position in Init Chain
 
-### 0. Project Brain Lookup (CHẠY TRƯỚC BEADS)
-
-**Mục đích:** Load project structure vào context — AI không cần scan cấu trúc lần nào nữa trong session này.
-
-```bash
-# Check project identity
-cat .project-identity 2>/dev/null || echo "No .project-identity found"
-
-# Check codebase map
-cat CODEBASE.md 2>/dev/null || echo "No CODEBASE.md found"
 ```
-
-**Output nếu tìm thấy:**
-```
-📚 [ProjectName] | [Stage] | [Architecture]
-🗺️  Codebase loaded: [N layers, M features indexed]
+symphony-orchestrator  (Gate 0: Server health + project overview)
+  ↓
+awf-session-restore    ← BẠN ĐANG Ở ĐÂY (Gate 0.5: Silent context gather)
+  ↓
+nm-memory-sync         (Gate 1: Associative memory sync)
+  ↓
+symphony-enforcer      (Gate 2: Project → Brain → Task → Confirmation block)
 ```
 
-**Output nếu không có:**
-```
-📚 No project identity found — raw mode
-```
+> **Chỉ `symphony-enforcer` mới in ra console block cho user.**
+> `awf-session-restore` chạy NGẦM, không in gì.
 
 ---
 
-### 1. Multi-Source Context Check (vẠn MANDATORY)
+## ⛔ Strict Execution Order (MANDATORY)
 
-**Priority 1: Beads (Task State) — PHẢI CHẠY ĐẦU TIÊN**
-```bash
-bd list --status in_progress
-bd list --status open --limit 3
-```
-
-**Output:**
-- Tasks đang làm dở
-- Tasks blocked
-- Tasks ready to start
-
-**Priority 2: Brain (Knowledge & Plans)**
-```bash
-# Check active plans
-cat brain/active_plans.json
+> [!CAUTION]
+> PHẢI tuân thủ thứ tự dưới đây. Vi phạm thứ tự = rò rỉ bộ nhớ đa dự án.
+> TUYỆT ĐỐI KHÔNG gọi `nmem_context`, `nmem_recap`, hay bất kỳ MCP memory tool nào
+> TRƯỚC KHI hoàn thành Step 1 và Step 2.
 
-# Check recent memories
-ls -lt brain/ | head -5
-```
-
-**Output:**
-- Plan đang active
-- Phase hiện tại
-- Recent decisions/knowledge
+### Step 1: Fetch Project Identity (CHẠY ĐẦU TIÊN)
 
-**Priority 3: Git (Code State)**
 ```bash
-git status
-git log -1
+cat .project-identity 2>/dev/null || echo "NO_PROJECT"
 ```
 
-**Output:**
-- Files đang thay đổi
-- Commit gần nhất
-
----
+**Nếu tìm thấy:** Extract `projectId` và `projectName` từ JSON.
+**Nếu không:** Ghi nhận `raw mode` — các bước sau vẫn chạy nhưng không scope theo project.
 
-### 2. Context Synthesis
+### Step 2: Switch NeuralMemory Brain (CHẠY THỨ HAI)
 
-Kết hợp 4 nguồn để tạo summary:
+> [!IMPORTANT]
+> Step này PHẢI hoàn thành TRƯỚC KHI gọi bất kỳ `nmem_*` MCP tool nào.
 
-```markdown
-🧠 **SESSION RESTORED**
-
-📚 **Project Brain:**
-- Project: FitBite Witness | Stage 3 | Clean Architecture + MVVM
-- CODEBASE: Loaded (4 layers, 8 features indexed)
-
-📿 **Beads Context:**
-- In Progress: Task #123 "Implement Login API" (started 2h ago)
-- Ready: 3 tasks
-
-🧠 **Brain Context:**
-- Active Plan: Shopping Cart - Phase 02 (50%)
-- Last Save: 30 minutes ago
-
-📂 **Git Context:**
-- Changed Files: 3 files (src/api/auth/*.ts)
-- Last Commit: "feat: add user model" (1 hour ago)
-
-➡️ **SUGGESTED NEXT STEP:**
-Continue task #123? `/codeExpert` or `/code`
+Nếu Step 1 tìm thấy `projectId`:
+```bash
+# CLI command — ép NeuralMemory server đổi sang đúng brain
+nmem brain use <projectId>
 ```
 
----
-
-### 3. Smart Suggestions
+**Chờ xác nhận đổi brain thành công** rồi mới tiếp tục Step 3.
 
-Based on context, suggest appropriate action:
+### Step 3: Gather Context (song song — tất cả silent)
 
-**Case 1: Task In-Progress**
-```
-➡️ Tiếp tục task #123? 
-   `/codeExpert` (Fast) or `/code` (Guided)
-```
+Sau khi brain đã switch, thu thập 3 nguồn context **song song**:
 
-**Case 2: Task Blocked**
+#### 3a. Git/Code State
+```bash
+git status --short 2>/dev/null
+git log -1 --oneline 2>/dev/null
 ```
-⚠️ Task #125 bị block bởi #120
 
-➡️ Làm task #120 trước?
-   `/codeExpert` or switch to another task
+#### 3b. Active Plans (scoped theo projectId)
+```bash
+# CHÚ Ý: Dùng projectId để scope đúng thư mục
+ls -t brain/<projectId>/*/implementation_plan.md 2>/dev/null | head -1
 ```
 
-**Case 3: No Active Task**
+Hoặc nếu dùng cấu trúc brain khác:
+```bash
+cat CODEBASE.md 2>/dev/null | head -5
 ```
-📋 Có 3 tasks ready to start
 
-➡️ Bắt đầu task mới?
-   `/next` để xem gợi ý
+#### 3c. Symphony Task State
+```bash
+symphony task list -P <projectId> -s in_progress --json 2>/dev/null
 ```
 
-**Case 4: Fresh Start**
+Hoặc dùng MCP tool (sau khi brain đã switch):
 ```
-🆕 Chưa có context
-
-➡️ Bắt đầu dự án mới?
-   `/planExpert "Feature"` or `/brainstorm`
+symphony_available_tasks(filter="my")
 ```
 
----
-
-### 4. Memory Persistence
+### Step 4: Compose Silent Context
 
-Update session state:
+**KHÔNG in ra console.** Tổng hợp thành context object ngầm cho AI sử dụng nội bộ:
 
 ```json
-// brain/session.json
 {
-  "last_session": "2026-01-30T10:00:00Z",
-  "working_on": {
-    "feature": "Shopping Cart",
-    "plan_path": "plans/260130-1025-shopping-cart/",
-    "current_phase": "phase-02",
-    "current_task": {
-      "id": 123,
-      "name": "Implement Login API",
-      "status": "in_progress"
-    }
+  "project": {
+    "id": "<projectId>",
+    "name": "<projectName>",
+    "codebase_loaded": true
+  },
+  "git": {
+    "changed_files": ["file1.swift", "file2.swift"],
+    "last_commit": "feat: add auth module"
+  },
+  "symphony": {
+    "active_tasks": ["sym-XYZ"],
+    "ready_tasks_count": 3
   },
-  "context_sources": {
-    "beads": true,
-    "brain": true,
-    "git": true
+  "brain": {
+    "switched_to": "<projectId>",
+    "active_plan": "plans/260316-auth/implementation_plan.md"
   }
 }
 ```
 
----
-
-## Error Handling
-
-### Beads Unavailable
-```
-⚠️ Beads không khả dụng
-
-Fallback: Dùng Brain + Git context
-```
-
-### Brain Empty
-```
-⚠️ Brain chưa có context
-
-Gợi ý: `/plan` để tạo context mới
-```
-
-### All Sources Fail
-```
-❌ Không thể khôi phục context
-
-➡️ Bắt đầu lại:
-1. `/recap` để quét dự án
-2. `/plan` để tạo plan mới
-3. Kể cho em biết đang làm gì
-```
+AI dùng context này để:
+- Hiểu user đang code dở gì
+- Biết task nào đang in-progress
+- Gợi ý tiếp tục đúng chỗ
 
 ---
 
-## Integration with Workflows
-
-### Auto-Trigger in Workflows
-
-Các workflows tự động gọi session restore:
-
-```markdown
-# In /code workflow
-1. Check session.json
-2. If no context → Trigger awf-session-restore
-3. Resume from restored context
-```
-
-### Manual Trigger
-
-User có thể gọi thủ công:
+## Error Handling
 
-```bash
-/recap    # Alias for session restore
-```
+| Tình huống | Xử lý |
+|-----------|--------|
+| `.project-identity` không tồn tại | Raw mode — skip brain switch, vẫn thu thập Git + Symphony global |
+| `nmem brain use` fail | Warning log — tiếp tục với Git + Symphony context |
+| Symphony server down | Đã được `symphony-orchestrator` xử lý trước đó — nếu vẫn down thì skip |
+| Git không phải repo | Skip git context — vẫn có Symphony + Brain |
+| Tất cả fail | AI bắt đầu với clean state — không block workflow |
 
 ---
 
-## Performance
+## What Changed from v6.4
 
-- **Execution Time:** < 1 second
-- **Sources Checked:** 3 (Beads, Brain, Git)
-- **Output:** Concise summary (< 10 lines)
+| v6.4 (Old) | v7.0 (New) |
+|------------|------------|
+| CLI task list | `symphony task list` (Symphony MCP) |
+| In block `🧠 SESSION RESTORED` | **Silent** — không in gì |
+| Đọc `.project-identity` và `CODEBASE.md` | Chỉ đọc `.project-identity` (CODEBASE gate do orchestrator xử lý) |
+| `brain/*` wildcard scan | `brain/<projectId>/` scoped access |
+| Không switch brain trước khi đọc memory | **BẮT BUỘC** `nmem brain use` trước mọi memory read |
+| Smart Suggestions block | Nhường cho `symphony next` |
+| 3 sources: CLI + Brain + Git | 3 sources: Symphony + Brain + Git |
 
 ---
 
-## Example Output
-
-```
-🧠 **WELCOME BACK!**
-
-📿 **Beads:** Task #123 "Login API" (in_progress, 2h ago)
-🧠 **Brain:** Plan "Shopping Cart" - Phase 02 (50%)
-📂 **Git:** 3 files changed
-
-➡️ **NEXT:** Continue coding? `/codeExpert`
+## Integration Notes
 
-💡 **TIP:** Gõ `/next` để xem chi tiết hơn
-```
+- **symphony-orchestrator** đã check server health → skill này KHÔNG cần check lại.
+- **symphony-enforcer** sẽ in confirmation block → skill này KHÔNG in gì.
+- **nm-memory-sync** chạy SAU skill này → brain đã switch đúng project.
+- **orchestrator** sẽ dispatch workflow → skill này chỉ gather context.
 
 ---
 
@@ -263,8 +185,10 @@ User có thể customize trong `brain/preferences.json`:
 {
   "session_restore": {
     "auto_trigger": true,
-    "verbosity": "concise",  // concise | detailed
-    "sources": ["beads", "brain", "git"]
+    "sources": ["symphony", "brain", "git"]
   }
 }
 ```
+
+> **Removed:** `verbosity` option (no longer needed — always silent).
+> **Removed:** Legacy CLI task source (replaced by Symphony MCP).

package/skills/beads-manager/SKILL.md

@@ -1,8 +1,8 @@
 ---
 name: beads-manager
-description: Smart Beads task management with Brain integration
+description: "⚠️ DEPRECATED — Replaced by symphony-orchestrator. Do NOT use this skill."
 version: 6.6.0
-trigger: always
+trigger: never
 invocation-type: auto
 ---
 

package/skills/brainstorm-agent/SKILL.md

@@ -30,7 +30,7 @@ priority: medium
 | Skill này LÀM | Skill này KHÔNG làm |
 |---------------|---------------------|
 | Brainstorm ý tưởng tính năng | Đọc/ghi brain/memory files |
-| Tư vấn hướng đi sản phẩm | Track tasks (việc của beads-manager) |
+| Tư vấn hướng đi sản phẩm | Track tasks (việc của symphony-orchestrator) |
 | Research thị trường | Sửa lỗi code |
 | Tạo BRIEF.md output | Lên kế hoạch chi tiết (việc của /plan) |
 | Phân tích đối thủ | Deploy |
@@ -249,6 +249,51 @@ Anh muốn làm gì tiếp:
 
 ---
 
+## 💾 SYMPHONY NOTES AUTO-SAVE (BẮT BUỘC)
+
+> **Rule:** Sau khi tạo BẤT KỲ brainstorm artifact nào (BRIEF.md, analysis, brainstorm doc),
+> AI PHẢI POST metadata vào Symphony Notes API. Content ĐỌC TỪ FILE, KHÔNG lưu vào DB.
+
+### Khi nào trigger:
+
+```yaml
+triggers:
+  - BRIEF.md tạo xong
+  - Analysis document tạo xong (pricing, market, competitor...)
+  - Brainstorm document tạo xong
+  - Bất kỳ artifact có giá trị lâu dài
+```
+
+### Cách thực hiện:
+
+```bash
+curl -X POST http://localhost:3100/api/notes -H 'Content-Type: application/json' -d '{
+  "projectId": "<current-project-id>",
+  "type": "brainstorm",
+  "title": "<artifact-title>",
+  "content": "<summary-2-3-lines-ONLY — KHÔNG copy full content>",
+  "filePath": "<absolute-path-to-artifact-file>",
+  "conversationId": "<current-conversation-id>",
+  "metadata": {
+    "mode": "quick|full|feature",
+    "tags": ["pricing", "features", "architecture", ...],
+    "created_by": "brainstorm-agent"
+  }
+}'
+```
+
+### Quy tắc:
+
+```yaml
+rules:
+  - content CHỈ chứa summary ngắn 2-3 dòng
+  - filePath trỏ đến file .resolved hoặc .md thực tế
+  - Nếu Symphony server không chạy → skip (không block workflow)
+  - type: "brainstorm" cho ý tưởng, "analysis" cho phân tích dữ liệu
+```
+
+---
+
 ## 🚫 ANTI-PATTERNS
 
 ```yaml
@@ -285,7 +330,7 @@ Short messages — không dump wall of text
 Works WITH:  /brainstorm workflow (skill này hỗ trợ workflow)
 Delegates TO: /plan (sau khi BRIEF xong)
 NOT: memory-sync (hoàn toàn độc lập — memory-sync tự theo dõi)
-NOT: beads-manager (không tạo task, chỉ brainstorm)
+NOT: symphony-orchestrator (không tạo task, chỉ brainstorm)
 Triggers: memory-sync W3 sẽ tự kích hoạt khi BRIEF.md tạo xong
 ```
 

package/skills/gemini-conductor/SKILL.md

@@ -0,0 +1,234 @@
+---
+name: gemini-conductor
+description: >-
+  Two-Agent Flow — Antigravity proactively invokes Gemini CLI (headless)
+  for project-wide analysis, strategic planning, and second opinions.
+  CLI runs on separate quota pool, extending AI capacity.
+metadata:
+  stage: core
+  version: "1.0"
+  requires: gemini (npm i -g @anthropic-ai/gemini-cli)
+  tags: [conductor, cli, two-agent, strategy, analysis, delegation]
+agent: Conductor
+trigger: conditional
+invocation-type: auto
+priority: 5
+---
+
+# 🎼 Gemini Conductor Skill
+
+> **Purpose:** Antigravity tự gọi Gemini CLI qua terminal khi cần tầm nhìn rộng hơn scope IDE.
+> **Key Benefit:** CLI dùng quota pool riêng → nhân đôi AI capacity. Giảm tunnel vision.
+
+---
+
+## ⚠️ Core Principle
+
+```
+Antigravity (IDE) = Executor — code, debug, file edits
+Gemini CLI       = Conductor — analysis, strategy, review
+
+Antigravity CHỦ ĐỘNG gọi CLI khi cần. User KHÔNG cần tự chuyển.
+```
+
+---
+
+## 🎯 Trigger Conditions
+
+Skill này kích hoạt khi Antigravity nhận diện task cần tầm nhìn rộng:
+
+```yaml
+auto_trigger:
+  high_confidence:
+    - Project-wide refactoring (>5 files affected)
+    - Architecture analysis / design review
+    - Cross-module dependency analysis
+    - Strategic planning for feature spanning multiple areas
+    - Need second opinion on complex technical decision
+
+  medium_confidence (confirm before invoking):
+    - Code review before commit (>3 files changed)
+    - Test plan generation for new feature
+    - Performance audit across codebase
+
+  never_trigger:
+    - Simple file edits (<3 files)
+    - Bug fix in single module
+    - UI changes in one screen
+    - Questions user can answer directly
+```
+
+---
+
+## 🔧 CLI Invocation Pattern
+
+### Base Command
+
+```bash
+gemini -p "ONLY edit/create .md files. DO NOT modify any code. <PROMPT>" \
+  --approval-mode auto \
+  -o json \
+  2>/dev/null
+```
+
+| Flag | Purpose |
+|------|---------|
+| `-p "..."` | Headless mode — MUST explicitly forbid editing source code. |
+| `--approval-mode auto` | Allows CLI to execute file edits (only `.md`), bypassing prompts |
+| `-o json` | Structured output for easy parsing |
+
+### Safety Rules
+
+```yaml
+safety:
+  - CLI CHỈ ĐƯỢC PHÉP tạo/sửa file tài liệu (`.md`). TUYỆT ĐỐI KHÔNG ĐƯỢC sửa code thực tế.
+  - ALWAYS inject explicit file-editing restrictions into the `-p` prompt.
+  - Timeout: 60s max per CLI call
+  - If CLI fails → gracefully fallback to Antigravity-only mode
+  - NEVER pass secrets/tokens in -p prompt
+  - Working directory: ALWAYS set to project root
+```
+
+### Command Template
+
+```bash
+# Safe invocation with timeout
+timeout 60 gemini -p "<prompt>. ONLY edit/create .md files. DO NOT touch source code." --approval-mode auto -o json 2>/dev/null
+```
+
+---
+
+## 📋 Use Cases & Prompt Templates
+
+### 1. Project Structure Analysis
+
+```bash
+gemini -p "Analyze the project structure in the current directory. \
+List main modules, their responsibilities, and key dependencies. \
+Output as structured JSON with modules array." \
+--approval-mode plan -o json
+```
+
+**When:** Starting work on unfamiliar part of codebase, or after long gap.
+
+### 2. Cross-Module Impact Analysis
+
+```bash
+gemini -p "I plan to modify <FILE_OR_MODULE>. \
+Analyze which other files/modules depend on it and would be impacted. \
+List files with risk level (high/medium/low)." \
+--approval-mode plan -o json
+```
+
+**When:** Before refactoring that touches shared interfaces.
+
+### 3. Strategic Refactoring Plan
+
+```bash
+gemini -p "Review the codebase and propose a refactoring strategy for <AREA>. \
+Consider: current architecture, dependencies, risk, and migration path. \
+Prioritize changes by impact and difficulty." \
+--approval-mode plan -o json
+```
+
+**When:** Large-scale refactoring spanning multiple modules.
+
+### 4. Second Opinion / Decision Support
+
+```bash
+gemini -p "Evaluate two approaches for <PROBLEM>: \
+Approach A: <desc>. Approach B: <desc>. \
+Compare: complexity, maintainability, performance, and risk. \
+Recommend one with reasoning." \
+--approval-mode plan -o json
+```
+
+**When:** Facing complex architectural decision.
+
+### 5. Pre-Commit Code Review
+
+```bash
+gemini -p "Review the following code changes for potential issues: \
+<DIFF_OR_DESCRIPTION>. \
+Check for: bugs, security issues, performance problems, and best practices." \
+--approval-mode plan -o json
+```
+
+**When:** Before committing changes across >3 files.
+
+### 6. Test Strategy Generation
+
+```bash
+gemini -p "Generate a comprehensive test strategy for <FEATURE>. \
+Include: unit tests, integration tests, edge cases, and test data. \
+Consider existing test patterns in the project." \
+--approval-mode plan -o json
+```
+
+**When:** Building new feature that needs thorough test coverage.
+
+---
+
+## 🔄 Integration Flow
+
+```
+1. Antigravity detects trigger condition
+2. Build CLI prompt with relevant context
+3. Run: run_command("timeout 60 gemini -p '...' --approval-mode plan -o json")
+4. Parse CLI output (JSON or text)
+5. Integrate insights into current task
+6. Continue with code execution using enriched context
+```
+
+### Output Handling
+
+```yaml
+on_success:
+  - Parse JSON output from CLI
+  - Extract key insights, recommendations, file lists
+  - Use them to guide next code edits
+  - Optionally save analysis to conductor/tracks.md
+
+on_timeout:
+  - Log: "⏳ CLI analysis timed out, proceeding with local context"
+  - Fall back to Antigravity-only analysis
+
+on_error:
+  - Log: "⚠️ CLI invocation failed, continuing without conductor"
+  - Do NOT block the workflow — CLI is enhancement, not dependency
+```
+
+---
+
+## 🚫 Anti-Patterns
+
+```yaml
+never_do:
+  - Call CLI for simple single-file edits
+  - Let CLI edit actual source code files (ONLY .md docs are allowed)
+  - Pass sensitive data (API keys, tokens) in prompts
+  - Block on CLI response indefinitely (always use timeout)
+  - Call CLI more than 3 times per task (diminishing returns)
+  - Ignore CLI output — if you called it, use the result
+
+always_do:
+  - Mention to user when invoking CLI: "📡 Đang gọi Gemini CLI phân tích..."
+  - Include project-specific context in CLI prompt
+  - Summarize CLI findings before acting on them
+  - Fall back gracefully if CLI unavailable
+```
+
+---
+
+## 🧩 Skill Relationships
+
+```
+Uses:      run_command (to invoke gemini CLI)
+Enhances:  /plan, /code, /debug, /refactor workflows
+Saves to:  conductor/tracks.md (optional)
+Independent of: NeuralMemory (CLI has its own context)
+```
+
+---
+
+*gemini-conductor v1.0 — Two-Agent Flow Skill for Antigravity*