Kinhluan Router

Kinhluan Router

Central dispatcher for the kinhluan skills ecosystem. Routes user requests to the correct skill based on workflow stage detection, artifact existence, and intent classification.

"The right tool for the right job at the right time."

0. Agent Charter — Expertise & Authority

0.1 Agent Identity

When activated, the agent must declare its expertise domain and current role:

"I am acting as a [Domain Expert] in [Specific Field]. 
 My objective is to [Task Goal] using the [Skill Name] framework.
 I will follow the Master Framework's [Layer X] workflow."

Expertise domains by skill:

0.2 User Sovereignty — The Golden Rule

┌─────────────────────────────────────────────────────────────┐
│  AGENT PROPOSES  →  USER DECIDES  →  AGENT EXECUTES         │
│                                                             │
│  • Agent NEVER makes irreversible decisions without consent │
│  • Agent ALWAYS presents trade-offs, not ultimatums         │
│  • User can OVERRIDE any recommendation at any time         │
│  • Agent must CONFIRM before: skip steps, change scope,     │
│    deviate from Master Framework                            │
└─────────────────────────────────────────────────────────────┘

Mandatory escalation points:

• Skip a workflow stage → Ask user: "Bước [X] chưa hoàn thành. Bạn có muốn bỏ qua không?"
• Change JTBD/Value Prop after PMR → Ask user: "PMR kết quả [X]. Có cập nhật JTBD không?"
• Override kill criteria → Ask user: "Kill criteria triggered: [reason]. Proceed anyway?"
• Select tech stack → Ask user: "Recommend [stack]. Approve or choose alternative?"

0.3 Master Framework Compliance

Before ANY recommendation, agent must check:

□ 1. What workflow stage is the user in? (PhD / MVP / Ngũ Sự / Tech)
□ 2. What artifacts already exist in workspace?
□ 3. What is the NEXT required step per Master Framework?
□ 4. Does user's request match the current stage?
□ 5. If NOT match → Escalate: "Bạn đang ở bước [X]. 
    Muốn làm [Y] trước, hay hoàn thành [X] trước?"

Compliance violation examples:

1. When to Use

• User says: "dùng skill phù hợp", "use right skill", "kinhluan skills", "what should I do next?"
• Multiple skills could apply to a request
• Unclear which workflow stage the user is in
• Need to check artifact existence before routing
• Cross-workflow routing (e.g., "write paper" but experiments not done yet)

2. Routing Architecture

2.1 Three-Layer Detection

Layer 1: WORKFLOW DETECTION
    ├─ PhD/Research workflow?
    ├─ MVP/Product workflow?
    ├─ Ngũ Sự/Strategy audit?
    └─ Pure tech/dev task?

Layer 2: STAGE / ARTIFACT DETECTION
    ├─ What files exist in workspace?
    ├─ What was the last completed step?
    └─ What gaps exist in the current workflow?

Layer 3: INTENT CLASSIFICATION
    ├─ Create / Write / Review / Analyze / Fix
    ├─ Specific domain (scheduling, FL, security...)
    └─ Urgency / priority signals

2.2 Workflow Detection Rules

3. PhD/Research Workflow Routing

Stage Map

Stage 0: START ──→ sota-survey (if no sota-matrix.md)
   │
Stage 1: SURVEY ──→ research-question (if sota-matrix.md exists, no RQ defined)
   │
Stage 2: RQ ──→ research-design (if RQ exists, no protocol)
   │
Stage 3: DESIGN ──→ experiment-tracking (if protocol exists, no experiments run)
   │
Stage 4: EXPERIMENT ──→ paper-writing (if results exist, no draft)
   │
Stage 5: DRAFT ──→ internal-critique → journal-q1-polish (if draft exists)
   │
Stage 6: POLISH ──→ publication-strategy (if paper ready)
   │
Stage 7: SUBMIT ──→ defense-prep (if paper accepted / thesis compilation)

Artifact-Based Routing

Cross-Stage Routing Examples

User: "viết paper về federated learning"

1. Detect: PhD workflow
2. Check: Does sota-matrix.md exist? → No
3. Route: "Bạn cần survey SOTA trước. Dùng sota-survey để tìm papers, 
   sau đó mới viết paper."

User: "review my paper"

1. Detect: PhD workflow, Stage 5-6
2. Check: Does draft exist? → Yes
3. Check: Has internal-critique been run? → No
4. Route: internal-critique
5. After critique: "Tiếp theo dùng journal-q1-polish để chuẩn hóa Q1."

User: "so sánh baseline"

1. Detect: PhD workflow, Stage 4
2. Check: experiments/ exists? → Yes
3. Route: experiment-tracking (comparison workflow)

4. MVP/Product Workflow Routing

Stage Map

Stage 0: START ──→ problem-discovery (if no problem statement)
   │
Stage 1: DISCOVERY ──→ why-strategic-rationale (if problem validated)
   │
Stage 2: WHY ──→ business-product-leadership (if WHY statement exists)
   │
Stage 3: WHAT ──→ ddd-core + c4-model (if JTBD defined)
   │
Stage 4: DESIGN ──→ ddd-tactical + c4-level2/3 (if bounded contexts defined)
   │
Stage 5: BUILD ──→ python-dev / js-dev / go-dev (if architecture ready)
   │
Stage 6: SHIP ──→ docker + k8s + dora-core + collaborative-engineering-agent
   │
Stage 7: RELEASE ──→ diffusion-release-tracking

Artifact-Based Routing

Ngũ Sự Audit Routing

At any stage, user can trigger Ngũ Sự assessment:

User: "đánh giá lại chiến lược" / "should we continue?"
→ art-of-war-software-engineering (score all 5 factors)
   ├─ Tao < 5 → why-strategic-rationale
   ├─ Heaven < 5 → problem-discovery + diffusion-release-tracking
   ├─ Earth < 5 → c4-model audit
   ├─ Command < 5 → business-product-leadership (DRI/RACI)
   └─ Method < 5 → dora-core baseline

5. Tech/Dev Task Routing

Language Detection

Task Type Detection

6. Domain-Specific Routing

These skills are domain-specific and only route when the domain is explicitly mentioned:

Rule: Do NOT route to domain-specific skills unless the user explicitly mentions the domain or the workspace contains domain-specific code (e.g., federated/, scheduling/ directories).

7. Multi-Skill Scenarios

Scenario: "Làm MVP app học tiếng Anh"

1. Workflow: MVP/Product
2. Check artifacts: None → Stage 0
3. Route sequence:
   a. problem-discovery → validate demand
   b. why-strategic-rationale → WHY statement
   c. business-product-leadership → JTBD + MVP scope
   d. ddd-core + c4-model → architecture
   e. js-dev / python-dev → build
   f. docker + k8s → ship
   g. diffusion-release-tracking → release

Scenario: "Viết luận văn thạc sĩ về optimization"

1. Workflow: PhD/Research
2. Check artifacts: None → Stage 0
3. Route sequence:
   a. sota-survey → literature matrix
   b. research-question → RQ + hypothesis
   c. research-design → protocol
   d. experiment-tracking → run experiments
   e. paper-writing → draft
   f. internal-critique → self-review
   g. journal-q1-polish → Q1 polish
   h. publication-strategy → venue + submit
   i. defense-prep → prepare defense

Scenario: "Review code Go microservice"

1. Workflow: Tech/Dev
2. Language: Go (go.mod detected)
3. Intent: code review
4. Route:
   a. golang-development (language patterns)
   b. collaborative-engineering-agent (DRE critique)
   c. security-analysis (if auth/API involved)

8. Quick Reference: Skill → Trigger Mapping

9. Integration with Other Skills