@leejungkiin/awkit 1.5.5 → 1.6.0

package/skill-packs/threejs/skills/threejs-webxr/SKILL.md

@@ -0,0 +1,104 @@
+---
+name: threejs-webxr
+description: "WebXR integration for three.js: WebXRManager and XRManager on the renderer, session initialization patterns, VRButton and ARButton helpers, XRControllerModelFactory and hand model families, XREstimatedLight, XRPlanes, and related addon Webxr utilities. Use for immersive sessions and controller/hand tracking—not for standard desktop camera projection (threejs-camera) or composer post effects (threejs-postprocessing)."
+---
+
+## When to use this skill
+
+**ALWAYS use this skill when the user mentions:**
+
+- Entering VR/AR, `navigator.xr`, reference spaces, session `requestAnimationFrame` loop via renderer
+- `VRButton`, `ARButton`, `XRButton` creation patterns from examples
+- Controller models, hand tracking meshes, estimated real-world lighting probes
+
+**IMPORTANT: webxr vs renderers vs camera**
+
+| Topic | Skill |
+|-------|--------|
+| Enable XR on renderer, sizing | **threejs-renderers** (basics) + **threejs-webxr** (session) |
+| Desktop projection | **threejs-camera** |
+
+**Trigger phrases include:**
+
+- "WebXR", "VRButton", "ARButton", "XRControllerModelFactory", "hand tracking"
+- "虚拟现实", "增强现实", "沉浸式"
+
+## How to use this skill
+
+1. **HTTPS** requirement and feature detection for XR availability.
+2. **Buttons**: use official button factories to create DOM entry points; handle session end.
+3. **Renderer**: call `renderer.xr.enabled = true` patterns per docs; prefer `setAnimationLoop` for XR loops.
+4. **Controllers**: attach models via factories; read gamepad axes/buttons carefully with fallbacks.
+5. **Hands**: opt-in hand models when runtime supports; performance implications.
+6. **Lighting**: `XREstimatedLight` for AR realism—combine with **threejs-lights** cautiously.
+7. **Exit**: restore non-XR render loop and resize handling on session end.
+
+### Example: VR session with error handling
+
+```javascript
+import { VRButton } from 'three/addons/webxr/VRButton.js';
+
+// Feature detection and session start
+if ('xr' in navigator) {
+  renderer.xr.enabled = true;
+  document.body.appendChild(VRButton.createButton(renderer));
+
+  renderer.xr.addEventListener('sessionstart', () => {
+    console.log('XR session started');
+  });
+  renderer.xr.addEventListener('sessionend', () => {
+    console.log('XR session ended — restoring desktop view');
+  });
+} else {
+  console.warn('WebXR not supported in this browser');
+}
+```
+
+See [examples/workflow-xr-button.md](examples/workflow-xr-button.md).
+
+## Doc map (official)
+
+| Docs section | Representative links |
+|--------------|----------------------|
+| Renderer XR | https://threejs.org/docs/WebXRManager.html |
+| Webxr addons | https://threejs.org/docs/VRButton.html |
+| Webxr addons | https://threejs.org/docs/ARButton.html |
+| Webxr addons | https://threejs.org/docs/XRControllerModelFactory.html |
+
+## Scope
+
+- **In scope:** Documented WebXR manager + listed addons for buttons/controllers/hands/planes.
+- **Out of scope:** Store submission, OpenXR runtime specifics, custom native layers.
+
+## Common pitfalls and best practices
+
+- Desktop testing requires XR emulation or hardware; fail gracefully.
+- Mismatched reference space causes floor offset—validate stage vs local-floor.
+- Heavy post chains may not meet VR frame time—profile aggressively.
+
+## Documentation and version
+
+WebXR entry points span **Addons → Webxr** and renderer [`WebXRManager`](https://threejs.org/docs/WebXRManager.html) in [three.js docs](https://threejs.org/docs/). Browser and device capabilities vary—answers should cite the official example name and three.js version when possible.
+
+## Agent response checklist
+
+When answering under this skill, prefer responses that:
+
+1. Link `WebXRManager`, `VRButton`, `ARButton`, or controller factories as relevant.
+2. Use `setAnimationLoop` patterns with **threejs-renderers** for XR render loops.
+3. Avoid duplicating desktop **threejs-camera** projection advice for XR eyes.
+4. Mention reference space choice (local-floor, etc.) at a high level with docs link.
+5. Flag performance interaction with **threejs-postprocessing** in VR.
+
+## References
+
+- https://threejs.org/docs/WebXRManager.html
+- https://threejs.org/docs/VRButton.html
+- https://threejs.org/docs/ARButton.html
+- https://threejs.org/docs/XRControllerModelFactory.html
+
+## Keywords
+
+**English:** webxr, vr, ar, xr session, controller, hand tracking, three.js
+
+**中文：** WebXR、VR、AR、VRButton、手柄、手部追踪、沉浸式、three.js

package/skill-packs/threejs/skills/threejs-webxr/examples/workflow-xr-button.md

@@ -0,0 +1,15 @@
+# Workflow: VR entry button
+
+## Steps
+
+1. Ensure site served over HTTPS and `navigator.xr` checked.
+
+2. `document.body.appendChild(VRButton.createButton(renderer));` per [VRButton](https://threejs.org/docs/VRButton.html) for your three version.
+
+3. `renderer.xr.enabled = true;` Use `renderer.setAnimationLoop` callback for render + updates.
+
+4. Handle session end to restore window `resize` behavior if needed.
+
+5. For AR passthrough use `ARButton` instead—see **threejs-webxr** doc map.
+
+Always verify against the official example that matches your three.js revision.

package/skills/CATALOG.md

@@ -1,7 +1,7 @@
 # 📋 AWF Skill Catalog
 
-> Classification of core runtime skills and optional packs.
-> Updated: 2026-04-02
+> Classification of all skills by type, trigger, and priority.
+> Updated: 2026-03-06
 
 ---
 
@@ -9,65 +9,60 @@
 
 | Type | Description | Example |
 |------|-------------|---------|
-| `auto` | Always active, runs silently without user command | `orchestrator`, `nm-memory-sync` |
+| `auto` | Always active, runs silently without user command | `orchestrator`, `memory-sync` |
 | `manual` | User must explicitly invoke via `/command` or keyword | `brainstorm-agent`, `ios-engineer` |
 
 ---
 
-## Default Runtime Profile (`awkit install`)
+## Active Skills
+
+| # | Skill | Type | Trigger | Priority | Version | Status |
+|---|-------|------|---------|----------|---------|--------|
+| 1 | `orchestrator` | `auto` | Always (first) | 1 | 2.1.0 | ✅ Active |
+| 2 | `awf-session-restore` | `auto` | Session start | 2 | — | ✅ Active |
+| 3 | `memory-sync` | `auto` | Always | 3 | 2.2.0 | ✅ Active |
+| 4 | `symphony-orchestrator` | `auto` | Always | 4 | 1.0.0 | ✅ Active |
+| 5 | `brainstorm-agent` | `manual` | `/brainstorm`, keywords | 5 | 1.0.0 | ✅ Active |
+| 6 | `awf-error-translator` | `auto` | Error detected | 6 | — | ✅ Active |
+| 7 | `awf-adaptive-language` | `auto` | Always | 7 | — | ✅ Active |
+| 8 | `ios-engineer` | `manual` | iOS tasks | — | — | ✅ Active |
+| 9 | `smali-to-kotlin` | `manual` | `/reverse-android` | 8 | — | ✅ Active |
+| 10 | `smali-to-swift` | `manual` | `/reverse-ios` | 9 | — | ✅ Active |
+| 11 | `awf-context-help` | `auto` | `/help`, stuck | — | — | ✅ Active |
+| 12 | `auto-save` | `auto` | Session end | — | — | ✅ Background |
+| 13 | `awf-version-tracker` | `auto` | Skill changes | — | — | ✅ Background |
+| 14 | `module-spec-writer` | `auto` | Gate 1.5 check fail | 1.5 | 1.0.0 | ✅ Active |
+| 15 | `spec-gate` | `auto` | Gate 2 check fail | 2 | 1.0.0 | ✅ Active |
+| 16 | `visual-design-gate` | `auto` | Gate 2.5 check fail | 2.5 | 1.0.0 | ✅ Active |
+| 17 | `trello-sync` | `auto` | Always | 2 | 3.0.0 | ✅ Active |
 
-`awkit install` now installs a lean runtime profile instead of the full source skill library.
-
-### Core Lifecycle & Coordination
-
-- `orchestrator`
-- `symphony-orchestrator`
-- `awf-session-restore`
-- `nm-memory-sync`
-- `symphony-enforcer`
-- `trello-sync`
-
-### Planning & Execution
-
-- `brainstorm-agent`
-- `module-spec-writer`
-- `spec-gate`
-- `visual-design-gate`
-- `single-flow-task-execution`
+---
 
-### Quality & Debugging
+## Quality & Discipline Skills (Superpowers-Inspired)
 
-- `systematic-debugging`
-- `verification-gate`
-- `code-review`
-- `gitnexus-intelligence`
+> Skills ported and adapted from [obra/superpowers](https://github.com/obra/superpowers) framework.
+> Integrated into AWKit with NeuralMemory + Symphony extensions.
 
-### Meta / Self-Evolution
+| # | Skill | Type | Trigger | Priority | Status |
+|---|-------|------|---------|----------|--------|
+| 14 | `verification-gate` | `auto` | Task completion, commit, deploy | 1 | ✅ Active |
+| 15 | `systematic-debugging` | `auto` | `/debug`, error detected, test failures | 2 | ✅ Active |
+| 16 | `code-review` | `auto` | Task completion, before merge | 3 | ✅ Active |
+| 17 | `writing-skills` | `manual` | Creating/modifying skills | — | ✅ Active |
 
-- `writing-skills`
-- `skill-creator`
-- `awf-version-tracker`
-- `auto-save`
-- `awf-adaptive-language`
-- `awf-context-help`
-- `awf-error-translator`
-- `gemini-conductor`
-- `codex-conductor`
-- `telegram-notify`
-- `ship-to-code`
 
 ---
 
-## Optional Packs
+## NeuralMemory Skill Pack (Optional Upgrade)
+
+When NeuralMemory is installed, these skills provide enhanced capabilities:
 
-| Pack | Purpose |
-|------|---------|
-| `mobile-ios` | iOS, SwiftUI, Xcode, reverse engineering |
-| `mobile-android` | Android, Kotlin, Compose, AdMob, APK analysis |
-| `marketing` | ASO, App Store marketing, analytics, growth |
-| `creator-studio` | Sprites, LucyLab TTS, short-form video workflows |
-| `neural-memory` | Memory audit, intake, evolution, extra workflows |
-| `superpowers` | Additional execution/TDD/reference skills ported from Superpowers |
+| # | Skill | Type | Replaces | Trigger |
+|---|-------|------|----------|---------|
+| 1 | `nm-memory-sync` | `auto` | `memory-sync` | Session start, debug, errors |
+| 2 | `nm-memory-intake` | `manual` | — | `/memory-intake` |
+| 3 | `nm-memory-audit` | `manual` | — | `/memory-audit` |
+| 4 | `nm-memory-evolution` | `manual` | — | `/memory-evolution` |
 
 ---
 
@@ -76,7 +71,7 @@
 Skills marked with self-evolution have a `## Learnings` section that accumulates insights:
 
 - ✅ `orchestrator` — routing improvements
-- ✅ `nm-memory-sync` — trigger pattern refinements
+- ✅ `memory-sync` — trigger pattern refinements
 - ✅ `symphony-orchestrator` — task management optimizations
 
 ---

package/skills/TRIGGER_INDEX.md

@@ -0,0 +1,30 @@
+# Skill Trigger Index
+
+> ⚠️ TRƯỚC mọi side-effect, scan bảng này → load skill tương ứng TRƯỚC khi act.
+> Check THEN act — never act THEN check.
+
+| Skill | Trigger | Path | Pri |
+|:---|:---|:---|:---|
+| orchestrator | Mọi request (routing) | orchestrator/SKILL.md | 🔴 |
+| symphony-enforcer | Task create/update/complete, progress report | symphony-enforcer/SKILL.md | 🔴 |
+| symphony-orchestrator | Session start, server check | symphony-orchestrator/SKILL.md | 🔴 |
+| awf-session-restore | Session start, init chain | awf-session-restore/SKILL.md | 🔴 |
+| nm-memory-sync | Session start/end, debug, errors | nm-memory-sync/SKILL.md | 🟡 |
+| verification-gate | Task completion, commit, deploy, success claims | verification-gate/SKILL.md | 🟡 |
+| code-review | Task completion, before merge | code-review/SKILL.md | 🟡 |
+| spec-gate | Gate 2 architecture design | spec-gate/SKILL.md | 🟡 |
+| brainstorm-agent | `/brainstorm`, ý tưởng, ideation | brainstorm-agent/SKILL.md | 🟢 |
+| skill-creator | `/create-agent-skill`, tạo/sửa skill | skill-creator/SKILL.md | 🟢 |
+| gitnexus-intelligence | Refactoring, impact analysis, pre-commit | gitnexus-intelligence/SKILL.md | 🟢 |
+| ship-to-code | Code porting, migration, language translation | ship-to-code/SKILL.md | 🟢 |
+| visual-design-gate | Gate 2.5 UI/UX design sync | visual-design-gate/SKILL.md | 🟢 |
+| module-spec-writer | Gate 1.5 COMPLEX + >3 modules | module-spec-writer/SKILL.md | 🟢 |
+
+## File Guards
+
+| File pattern | PHẢI đọc lại | Lý do |
+|:---|:---|:---|
+| `symphony/**/*.json` | `symphony-enforcer/SKILL.md` | Task integrity |
+| `.project-identity` | `orchestrator/SKILL.md` | Project config |
+| `**/SKILL.md` | `skill-creator/SKILL.md` | Skill format |
+| `GEMINI.md` | N/A (self-referencing) | Core rules |

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

@@ -15,7 +15,9 @@ invocation-type: auto
 priority: 2
 ---
 
-# AWF Session Restore (v7.0 — Symphony Native)
+<!-- ⚠️ INIT CHAIN CRITICAL — File này quyết định thứ tự context loading. Sai = rò rỉ cross-project memory. -->
+
+# AWF Session Restore (v7.2 — Symphony Native + Trigger Index)
 
 > **Purpose:** Silently gather unstructured context (Git, Plans, Memory) at session start.
 > **Key Change v7.0:** Symphony-native. Strict brain scoping.
@@ -63,7 +65,8 @@ cat .project-identity 2>/dev/null || echo "NO_PROJECT"
 ```
 
 **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.
+**Cache Mindful Config:** Extract `mindfulCheckpoint` và `mindfulCheckpointConfig` (defaults: enabled=true, threshold=3, scopeGuard=true, milestoneRest=true).
+**Nếu không:** Ghi nhận `raw mode` — các bước sau vẫn chạy nhưng không scope theo project. Mindful defaults vẫn ON.
 
 ### Step 2: Switch NeuralMemory Brain (CHẠY THỨ HAI)
 
@@ -119,6 +122,14 @@ Hoặc dùng MCP tool (sau khi brain đã switch):
 symphony_available_tasks(filter="my")
 ```
 
+#### 3e. Skill Trigger Index (v7.2 — Two-Tier Loading)
+```
+Read: ~/.gemini/antigravity/skills/TRIGGER_INDEX.md
+→ Memorize trigger table (~150 tokens)
+→ Load specific SKILL.md files ON DEMAND khi action matches trigger
+→ KHÔNG load toàn bộ SKILL.md files upfront
+```
+
 ### Step 4: Compose Silent Context
 
 **KHÔNG in ra console.** Tổng hợp thành context object ngầm cho AI sử dụng nội bộ:
@@ -141,6 +152,12 @@ symphony_available_tasks(filter="my")
   "brain": {
     "switched_to": "<projectId>",
     "active_plan": "plans/260316-auth/implementation_plan.md"
+  },
+  "mindful": {
+    "enabled": true,
+    "iteration_threshold": 3,
+    "scope_guard": true,
+    "milestone_rest": true
   }
 }
 ```

package/skills/brainstorm-agent/SKILL.md

@@ -17,6 +17,8 @@ activation_keywords:
 priority: medium
 ---
 
+<!-- ⚠️ GATE 1 — Brainstorm ONLY. Không code trong phase này. Output = BRIEF.md. -->
+
 # 💡 Brainstorm Agent — Router
 
 > **Purpose:** Biến ý tưởng mơ hồ thành bản thiết kế rõ ràng qua phiên brainstorm có cấu trúc.

package/skills/code-review/SKILL.md

@@ -3,6 +3,8 @@ name: code-review
 description: Use when completing tasks, implementing features, or before merging. Dispatch structured code review with severity classification. Auto-triggers after task completion in subagent-driven or single-flow execution.
 ---
 
+<!-- ⚠️ REVIEW GATE — Spec compliance FIRST, then code quality. No merge without review. -->
+
 # Code Review
 
 ## Overview

package/skills/code-review-system-principles/SKILL.md

@@ -0,0 +1,154 @@
+---
+name: code-review-system-principles
+description: "Nguyên tắc cốt lõi, tiêu chí đánh giá và output schema cho hệ thống tự động Code Review. Sử dụng khi AI đóng vai trò người kiểm duyệt (reviewer) tìm bugs trong một bản vá (patch/diff)."
+version: 1.0.0
+trigger: conditional
+activation_keywords:
+  - "code review principles"
+  - "review principles"
+  - "nguyên tắc review"
+  - "code review rules"
+---
+
+<!-- ⚠️ DUAL-SYNCED: Thực thi đồng bộ chéo giữa ~/.gemini/antigravity và main-awf. -->
+
+# Code Review System Principles
+
+This document summarizes the active review prompt rules used when evaluating a patch.
+It is a human-readable restatement, not a verbatim raw prompt dump.
+
+## 1. Reviewer role
+
+- Act as a reviewer for a proposed code change made by another engineer.
+- Focus on discrete, actionable bugs introduced by the patch.
+- Prefer no findings over weak or speculative findings.
+
+## 2. What counts as a bug worth flagging
+
+A finding should generally satisfy all of the following:
+
+1. It meaningfully impacts correctness, performance, security, or maintainability.
+2. It is discrete and actionable.
+3. Fixing it matches the rigor level already present in the codebase.
+4. It was introduced by the current patch, not pre-existing.
+5. The original author would likely want to fix it if informed.
+6. It does not depend on unstated assumptions about intent.
+7. It identifies provably affected code or scenarios, not vague downstream risk.
+8. It is not merely an intentional product or behavior change.
+
+## 3. What to ignore
+
+- Trivial style issues, unless they obscure meaning or violate a documented standard.
+- General codebase quality complaints.
+- Speculative regressions without a concrete affected path.
+- Pre-existing problems not introduced by the patch.
+- Non-actionable feedback.
+
+## 4. Finding coverage rules
+
+- Return **all** findings the author would likely fix.
+- Do **not** stop at the first valid issue.
+- If there is no clearly valuable issue, return an empty findings list.
+
+## 5. Comment-writing rules
+
+Each finding comment should:
+
+- Clearly explain why the issue is a bug.
+- Accurately communicate severity.
+- Stay brief: at most one paragraph.
+- Avoid unnecessary praise or blame.
+- State the conditions, environments, or inputs required for the issue to occur.
+- Be understandable without close reading.
+- Avoid large code blocks; any included code should be very small.
+- Use matter-of-fact, assistant-like tone.
+- Use an imperative finding title with a priority prefix.
+- Keep the title short (no more than 80 characters).
+
+## 6. Location rules
+
+- Use one comment per distinct issue.
+- Keep the line range as short as possible.
+- Prefer the smallest diff-overlapping range that pinpoints the issue.
+- Avoid large ranges when a smaller subrange can explain the bug.
+
+## 7. Suggestion block rules
+
+- Use `suggestion` blocks **only** for concrete replacement code.
+- Keep them minimal.
+- Preserve exact leading whitespace.
+- Do not change outer indentation unless the fix truly requires it.
+
+## 8. Priority rules
+
+Every finding title must begin with a priority tag:
+
+- `[P0]` Blocking / universal / drop-everything issue
+- `[P1]` Urgent, should be fixed next cycle
+- `[P2]` Normal bug, fix eventually
+- `[P3]` Low priority, nice to have
+
+Also include numeric priority in JSON:
+
+- `0` for P0
+- `1` for P1
+- `2` for P2
+- `3` for P3
+
+## 9. Output schema rules
+
+The output must be valid JSON with exactly this top-level shape:
+
+- `findings`: array
+- `overall_correctness`: `"patch is correct"` or `"patch is incorrect"`
+- `overall_explanation`: short justification
+- `overall_confidence_score`: float
+
+Each finding must include:
+
+- `title`
+- `body`
+- `confidence_score`
+- `priority` (when determinable)
+- `code_location.absolute_file_path`
+- `code_location.line_range.start`
+- `code_location.line_range.end`
+
+Additional formatting constraints:
+
+- Output raw JSON only; no markdown fences or extra prose.
+- `code_location` is required for every finding.
+- The location should overlap the diff.
+- Do not generate a PR fix in the final review output.
+
+## 10. Correctness verdict rules
+
+Use `patch is correct` only when:
+
+- Existing code and tests should not break, and
+- The patch is free of blocking bugs.
+
+Ignore non-blocking nits such as:
+
+- style
+- formatting
+- typos
+- documentation polish
+
+## 11. Practical review heuristic
+
+A good finding usually answers all of these quickly:
+
+- **Where is the problem?**
+- **When does it break?**
+- **Why is it a real bug?**
+- **How severe is it?**
+- **Is it definitely introduced by this patch?**
+
+## 12. Review philosophy summary
+
+- Precision over volume
+- Evidence over speculation
+- Patch-introduced regressions over legacy issues
+- Author-useful feedback over theoretical concerns
+- Concise comments over long essays

package/skills/codex-conductor/SKILL.md

@@ -15,6 +15,8 @@ invocation-type: auto
 priority: 5
 ---
 
+<!-- ⚠️ INSPECT ONLY — Codex TUYỆT ĐỐI KHÔNG sửa code. Output = .md reports. approval-mode suggest. -->
+
 # 🔍 Codex Conductor — Router
 
 > **Purpose:** Gọi Codex CLI qua terminal khi cần rà soát logic, debug, review code chuyên sâu.
@@ -38,7 +40,7 @@ Codex CLI         = Inspector — debug, review, verify, test
 
 ```bash
 which codex || command -v codex
-# NOTE: If not installed, IMMEDIATELY fallback to IDE context. DO NOT ask user to install.
+# If not installed: npm i -g @openai/codex
 ```
 
 ## 🎯 Trigger Conditions
@@ -88,9 +90,9 @@ cd <PROJECT_ROOT> && timeout 180 codex exec \
 | Event | Action |
 |-------|--------|
 | Success | Parse → save report → summarize → suggest fixes |
-| Timeout | "⏳ Timed out (>120s)" → fallback Antigravity-only smoothly |
-| Error | Fallback Antigravity-only smoothly. Do NOT ask user to install |
-| Not installed | Fallback Antigravity-only smoothly. Do NOT ask user to install |
+| Timeout | "⏳ Timed out (>120s)" → fallback Antigravity-only |
+| Error | Check `codex --version` → suggest install |
+| Not installed | Ask user to install |
 
 ## 📁 Report Structure
 

package/skills/gemini-conductor/SKILL.md

@@ -15,6 +15,8 @@ invocation-type: auto
 priority: 5
 ---
 
+<!-- ⚠️ READ-ONLY CLI — Gemini CLI KHÔNG ĐƯỢC sửa code. Chỉ .md files. Timeout 60s. -->
+
 # 🎼 Gemini Conductor Skill
 
 > **Purpose:** Antigravity tự gọi Gemini CLI qua terminal khi cần tầm nhìn rộng hơn scope IDE.
@@ -194,9 +196,8 @@ on_timeout:
   - Fall back to Antigravity-only analysis
 
 on_error:
-  - Log: "⚠️ CLI invocation failed or not set up, continuing without conductor"
-  - IMMEDIATELY fallback to Antigravity-only mode
-  - Do NOT ask the user to install or authenticate the CLI. It is strictly optional
+  - Log: "⚠️ CLI invocation failed, continuing without conductor"
+  - Do NOT block the workflow — CLI is enhancement, not dependency
 ```
 
 ---

package/skills/gitnexus-intelligence/SKILL.md

@@ -14,6 +14,8 @@ invocation-type: auto
 priority: 5
 ---
 
+<!-- ⚠️ IMPACT-FIRST — PHẢI chạy gitnexus_impact TRƯỚC khi edit symbol. Chưa index → KHÔNG dùng tools. -->
+
 # GitNexus Intelligence v1.0 — Code-Aware Agent Skill
 
 > **Purpose:** Cung cấp architectural awareness cho AI agent thông qua knowledge graph.

package/skills/module-spec-writer/SKILL.md

@@ -23,6 +23,8 @@ activation_keywords:
   - "product spec"
 ---
 
+<!-- ⚠️ GATE 1.5 — User PHẢI approve module specs. COMPLEX + >3 modules = MANDATORY. -->
+
 # Module Spec Writer v1.2 — Gate 1.5: Product Specification (Router)
 
 > **Purpose:** Tạo tài liệu product-level cho từng module (screens, flows, rules)

package/skills/nm-memory-sync/SKILL.md

@@ -187,6 +187,16 @@ If `nmem` is not available AND auto-install fails, falls back to flat-file behav
 - Shows warning once: "⚠️ NeuralMemory not installed. Using flat-file fallback."
 ```
 
+## CLI Fallback (MCP Unresponsive)
+
+If the `neural-memory` MCP server is unresponsive or fails, you MUST fallback to using the `nmem` CLI directly using the `run_command` tool. The CLI offers full parity with the MCP tools:
+- `nmem_remember` -> `nmem remember "..."`
+- `nmem_recall` -> `nmem recall "..."` 
+- `nmem_context` -> `nmem recap` or `nmem recall "..."`
+- `nmem_auto` -> `nmem auto process --text "..."`
+
+Always check the `nmem` CLI help (`nmem --help`) if unsure of the syntax.
+
 ---
 
 ## Integration

package/skills/skill-creator/SKILL.md

@@ -11,6 +11,8 @@ activation_keywords:
   - "bổ sung skill"
 ---
 
+<!-- ⚠️ DUAL-SYNC BẮT BUỘC — Mọi thay đổi PHẢI sync sang main-awf. Không được bỏ qua Bước 3. -->
+
 # 🛠️ Skill Creator (with Auto-Sync to main-awf)
 
 **Purpose**: Hỗ trợ người dùng thiết kế, lập trình và tinh chỉnh các Antigravity Skills theo chuẩn. 

package/skills/spec-gate/SKILL.md

@@ -21,6 +21,8 @@ activation_keywords:
   - "API design"
 ---
 
+<!-- ⚠️ GATE 2 — User PHẢI approve design. AI KHÔNG được tự approve. Đọc GEMINI.md § 7-Gate trước khi sửa. -->
+
 # Spec Gate v1.2 — Architecture & Data Design Gate (Router)
 
 > **Purpose:** Chốt thiết kế kỹ thuật (DB Schema, API Contract, State Machine)

package/skills/symphony-enforcer/SKILL.md

@@ -17,6 +17,8 @@ invocation-type: auto
 priority: 1
 ---
 
+<!-- ⚠️ TASK INTEGRITY — Sửa file này PHẢI đọc lại GEMINI.md § Symphony rules. Không sửa format Sync Block. -->
+
 # Symphony Enforcer v3.7 — Router
 
 > **Purpose:** Đảm bảo AI KHÔNG BAO GIỜ quên cập nhật Symphony.
@@ -31,6 +33,12 @@ KHÔNG CÓ NGOẠI LỆ:
 - AI tự detect completion và auto-complete task
 - Task done → PHẢI atomic git commit trước khi suggest next
 - BỎ QUA BẤT KỲ STEP NÀO = VI PHẠM
+
+GATE 3 VIOLATION (NẶNG):
+- Tạo implementation plan mà KHÔNG tạo Symphony tickets = VI PHẠM NẶNG
+- Dùng artifact task.md thay cho Symphony = VI PHẠM
+  (task.md = IDE-level tracking, KHÔNG thay thế Symphony)
+- Severity tương đương: code logic khi chưa confirm UI
 ```
 
 ## 📋 Topic Index — Load file theo nhu cầu
@@ -41,6 +49,7 @@ KHÔNG CÓ NGOẠI LỆ:
 | Three-Phase Execution | Gate 4 với COMPLEX + UI tasks | `examples/three-phase.md` |
 | Trigger Points (TP1, TP1.5, TP1.7) | Milestone, schema changes, checkpoints | `examples/trigger-points.md` |
 | Task Completion (TP2-TP4) | Khi task sắp done | `examples/task-completion.md` |
+| Mindful Stop (TP-ITER, TP-SCOPE, TP-REST) | Iteration counter, scope guard, milestone rest | `examples/mindful-stop.md` |
 
 ## Auto-Lifecycle: task_boundary ↔ Symphony