refactor: remove legacy CLI skills, dead code, simplify presets

Remove 14 legacy skill templates (architecture-understanding, code-discovery, etc.)
Remove dead code: init-wizard, skill-manager, template-selector, migrate-skills, config/index
Simplify presets and skill discovery
Update CLAUDE.md governance section and template

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: Dchuong03

CLAUDE.md

@@ -19,7 +19,7 @@ Keep this managed block so 'openspec update' can refresh the instructions.
 
 <!-- GOVERNANCE:START -->
 ## Governance
-Task lifecycle, quality gates, lesson detection: managed by `sidstack-aware` skill. Use `/sidstack-dev` for structured workflows. SidStack MCP tasks (`mcp__sidstack__task_*`) for persistence; built-in tasks for session-local sub-steps only.
+Task lifecycle, quality gates, lesson detection: managed by `sidstack-aware` skill. Every request is **classified by intent first** (via Workflow Router + sidstack-aware skill), then routed to the correct workflow. Use `/sidstack-dev` for explicit structured workflows. SidStack MCP tasks (`mcp__sidstack__task_*`) for persistence; built-in tasks for session-local sub-steps only.
 
 **Integrated workflow — always follow this pattern:**
 1. **Before work:** `knowledge_search` + `memory_search` → find context. `entity_link` → link relevant docs to task.
@@ -29,6 +29,71 @@ Task lifecycle, quality gates, lesson detection: managed by `sidstack-aware` ski
 
 # SidStack - Claude Code Instructions
 
+## SidStack Governance
+
+This project uses **SidStack** for task tracking, knowledge management, and quality enforcement.
+
+---
+
+## CRITICAL RULES
+
+**1. QUALITY BEFORE COMPLETE**
+- Run project's test/build commands before marking task done
+- Verify the change works at runtime (not just static checks)
+- Never mark completed if tests fail
+
+**2. ASK WHEN UNCLEAR**
+- Don't guess requirements - ask the user
+- Don't assume scope - clarify before expanding
+
+---
+
+## Session Start
+
+At the beginning of each session, check for active work:
+```
+task_list({ projectId: "sidstack", preset: "actionable" })
+```
+If there's an in_progress task, resume it. Otherwise, show pending tasks.
+
+---
+
+## Workflow Router
+
+Classify every user request before acting. Match **semantic intent**, not keywords.
+
+| User wants to... | Workflow | Action |
+|-----------------|----------|--------|
+| Ask, discuss, explore ideas | **discuss** | Answer directly. No task needed. |
+| Create, list, or check task status | **track** | `task_list` / `task_create` / `task_update` only |
+| Fix critical production issue NOW | **hotfix** | `task_create(bugfix, critical)` → `/sidstack-dev hotfix` |
+| Fix a bug, error, broken behavior | **bugfix** | `task_create(bugfix)` → `/sidstack-dev fix` |
+| Build new feature or capability | **implement** | `task_create(feature)` → `/sidstack-dev feature` |
+| Refactor, optimize, clean up code | **improve** | `task_create(refactor/perf/debt)` → `/sidstack-dev feature` |
+| Review code or audit existing work | **review** | `/sidstack-dev review` |
+| Process a ticket from queue | **ticket** | `/sidstack ticket <id>` |
+| Build or update documentation/knowledge | **knowledge** | `/sidstack-knowledge` |
+| Record lessons from incidents | **learn** | `incident_create` → `lesson_create` |
+
+### Classification Rules
+
+1. **Intent over keywords.** "Fix the tests" = bugfix. "Add tests" = implement. "Why do tests fail?" = discuss.
+2. **Urgency splits hotfix from bugfix.** Hotfix = production is broken NOW. Bugfix = known issue, can plan.
+3. **Any language.** Classify by meaning — works in Vietnamese, English, etc.
+4. **When ambiguous** — ask user: "This could be [X] or [Y], which fits?"
+5. **Explicit `/sidstack-dev` always wins** over this router.
+
+### Task Size → Flow Depth
+
+| Size | Criteria | Flow |
+|------|----------|------|
+| **Trivial** | Typo, config, <5 lines | No task. Just do it. |
+| **Small** | Bug fix, single-file | `task_create` → implement → `task_complete` |
+| **Medium** | Multi-file feature | `task_create` → `entity_context` → implement → test → `task_complete` |
+| **Large** | Architecture change | `/sidstack-dev feature` (full 4-step with plan review) |
+
+---
+
 ## Project Overview
 **SidStack** = AI-Powered Project Intelligence Platform
 Core: Knowledge System, Impact Analysis, Task Management, Ticket Queue, Training Room.
@@ -78,6 +143,77 @@ pnpm typecheck            # Type checking
 **Impact:** `impact_analyze` → `impact_check_gate`. Gates: `blocked`, `warning`, `clear`.
 **Tickets:** `new → reviewing → approved → in_progress → completed` (or `rejected`).
 
+---
+
+## SidStack Tools Reference
+
+| Workflow Moment | Tool | Purpose |
+|----------------|------|---------|
+| **Starting Work** | `task_list(preset: "actionable")` | See what needs doing |
+| | `entity_context(entityType: "task", entityId: taskId)` | Load ALL linked context in one call |
+| | `task_create(projectId, title, taskType)` | Create new task |
+| **During Work** | `task_update(taskId, progress: 0-100)` | Track progress |
+| | `impact_analyze(description, targetFiles)` | Before touching core/risky code |
+| | `knowledge_search(query)` | Find project patterns, docs |
+| **Finishing Work** | `task_complete(taskId)` | After quality gates pass |
+| | `memory_add(content, projectId)` | Store learnings for future |
+| | `lesson_create(...)` | After fixing tricky bugs (ask user first) |
+
+> **Tip:** `entity_context` loads everything linked to a task (knowledge, memory, references) in one call — use it instead of calling `knowledge_search` + `memory_search` + `entity_references` separately.
+
+---
+
+## Slash Commands
+
+| Command | Purpose |
+|---------|---------|
+| `/sidstack` | Dashboard - tasks, tickets, focus, governance |
+| `/sidstack:knowledge` | Knowledge system - search, explore, build context |
+| `/sidstack:agent` | Spawn governed agent (worker/reviewer) |
+
+---
+
+## Quality Checks
+
+Run before completing any task:
+```bash
+pnpm test
+pnpm packages:build
+pnpm typecheck
+```
+
+---
+
+## Common Mistakes to Avoid
+
+| Mistake | Consequence | Correct Approach |
+|---------|-------------|------------------|
+| Skip task_create for "quick fix" | No audit trail, no acceptance criteria | Always create task, even for small changes |
+| Mark complete before testing | Broken code merged | Run tests + verify runtime first |
+| Expand scope without asking | Wasted effort, wrong direction | Clarify with user before expanding |
+| Ignore existing knowledge | Reinvent patterns, miss constraints | Call `knowledge_search` before implementing |
+
+---
+
+## Task Systems: SidStack MCP vs Built-in
+
+| System | Use For |
+|--------|---------|
+| **SidStack MCP** (`mcp__sidstack__task_*`) | Governance, quality gates, cross-session persistence |
+| **Built-in** (`TaskCreate/TaskUpdate`) | Session-local sub-step coordination |
+
+Rule: Always create the SidStack MCP task first (for governance), then optionally use built-in tasks for sub-steps.
+
+---
+
+## Agent Teams
+
+For multi-agent orchestration with Worker + Reviewer teammates:
+- Use `/sidstack:agent` slash command to spawn governed agents
+- See `.sidstack/governance.md` for spawn templates and orchestration workflow
+
+---
+
 <!-- DOCUMENTATION-DISCIPLINE:START -->
 ## MANDATORY: Documentation Discipline
 
@@ -107,3 +243,12 @@ pnpm typecheck            # Type checking
 1. Add Neo4j, Qdrant, or Go service dependencies
 2. Skip quality gates
 3. Create tasks without analysis
+
+---
+
+## More Details
+
+For comprehensive governance documentation:
+- `.sidstack/governance.md` - Full governance rules, agent teams guide
+- `.claude/skills/` - Auto-triggered Claude Code skills
+- `.claude/agents/` - Role-specific agent definitions (Worker, Reviewer)