Build RAG-driven coaching systems using ADBPG Supabase (training data management) + ADBPG Vector Knowledge Base (RAG-driven intelligent coaching).

Core Capabilities:

• Supabase Project: PostgreSQL-based platform for managing coaching domains, learners, and session records
• Vector Knowledge Base: ADBPG instance with vector optimization for RAG-driven coaching
• Seamless Integration: Supabase stores structured data, ADBPG powers intelligent AI coaching dialogue

Pre-check: Aliyun CLI >= 3.3.1 Required

IMPORTANT: Run aliyun version to verify >= 3.3.1. If not installed or version too low, see references/cli-installation-guide.md.

Then [MUST] run aliyun configure set --auto-plugin-install true to enable automatic plugin installation.

Pre-check: Alibaba Cloud Credentials Required

Security Rules:
- NEVER read, echo, or print AK/SK values
- NEVER ask the user to input AK/SK directly
- NEVER print passwords or API Keys in plain text in logs or stdout
- ONLY use aliyun configure list to check credential status
- When displaying API Keys, show only the first 6 characters + (e.g., sk-abc1)

aliyun configure list

If no valid profile exists, STOP here. Configure credentials outside of this session via aliyun configure or environment variables.

Scenario Description

Architecture

User (Web / Terminal / Agent)
           │
    ┌──────┴──────┐
    v             v
┌─────────────┐  ┌────────────────────────┐
│  Supabase   │  │  Agent Mode            │
│  (spb-xxx)  │  │  ChatWithKnowledgeBase │
│  - Domains  │  └───────────┬────────────┘
│  - Sessions │              │
└──────┬──────┘              │
       v                     v
┌────────────────────────────────────────┐
│  ADBPG Instance (gp-xxx) + KB          │
│  Domain Knowledge + RAG + LLM          │
└────────────────────────────────────────┘

RAM Policy

Required Permissions

See references/ram-policies.md for complete list.

[MUST] Permission Failure Handling: When any command fails due to permission errors:
1. Read references/ram-policies.md for required permissions
2. Use ram-permission-diagnose skill to guide the user
3. Pause and wait until user confirms permissions granted

Core Workflow

When user says "Help me set up an AI coaching system" or similar, execute the following steps:

Smart Defaults Mode: User only needs minimal input (e.g., "北京i"). The agent auto-parses region, discovers VPC/VSwitch, generates passwords, and presents all parameters for one-click confirmation.

Step 1: Create Supabase Project

Parameters to confirm for this step:

| Parameter | Default | Notes |
|-----------|---------|-------|
| RegionId | Auto-parse | "北京i" → cn-beijing, "上海b" → cn-shanghai, "杭州" → cn-hangzhou, "深圳" → cn-shenzhen |
| ZoneId | Auto-parse | "北京i" → cn-beijing-i; query zones when only city provided |
| VpcId | Auto-discover | Query available VPCs, select one with most available IPs |
| VSwitchId | Auto-discover | Query VSwitches in target zone, select one with most available IPs |
| ProjectName | ai_coaching | Supabase project name |
| AccountPassword | Auto-generate | Password rules: 8-32 chars, at least 3 of uppercase/lowercase/digits/special (@#$%^&*), avoid ! |

1.1 Check/Create NAT Gateway

Important: Supabase public connection requires a NAT Gateway with SNAT rules in the VPC.

# Check existing NAT Gateways in VPC
aliyun vpc describe-nat-gateways --profile adbpg \
  --biz-region-id  --vpc-id  \
  --user-agent AlibabaCloud-Agent-Skills

• If TotalCount > 0 and SNAT entries cover the VSwitch CIDR → Skip to Step 1.2
• If no NAT Gateway → Get user confirmation, then:

# 1.1a: Get VSwitch CIDR
aliyun vpc describe-vswitch-attributes --profile adbpg \
  --biz-region-id  --vswitch-id  \
  --user-agent AlibabaCloud-Agent-Skills
# Record: CidrBlock
# 1.1b: Create Enhanced NAT Gateway (requires user confirmation)
# 💰 Cost note: NAT Gateway incurs hourly charges
aliyun vpc create-nat-gateway --profile adbpg \
  --biz-region-id  --vpc-id  --vswitch-id  \
  --nat-type Enhanced \
  --user-agent AlibabaCloud-Agent-Skills
# Record: NatGatewayId and SnatTableIds.SnatTableId[0]
# Poll until Status=Available
# 1.1c: Find or allocate EIP (requires user confirmation)
# 💰 Cost note: EIP incurs charges; release via VPC console when no longer needed
aliyun vpc describe-eip-addresses --profile adbpg \
  --biz-region-id  \
  --user-agent AlibabaCloud-Agent-Skills
# If no available EIP:
aliyun vpc allocate-eip-address --profile adbpg \
  --biz-region-id  \
  --user-agent AlibabaCloud-Agent-Skills
# Record: AllocationId and EipAddress
# 1.1d: Bindind EIP to NAT Gateway (requires user confirmation)
aliyun vpc associate-eip-address --profile adbpg \
  --biz-region-id  \
  --allocation-id  --instance-id  \
  --instance-type Nat \
  --user-agent AlibabaCloud-Agent-Skills
# 1.1e: Create SNAT entry (requires user confirmation)
aliyun vpc create-snat-entry --profile adbpg \
  --biz-region-id  \
  --snat-table-id  \
  --source-cidr "" --snat-ip "" \
  --user-agent AlibabaCloud-Agent-Skills

1.2 Create Supabase Project

aliyun gpdb create-supabase-project --profile adbpg \
  --biz-region-id  --zone-id  \
  --project-name  --account-password '' \
  --security-ip-list "127.0.0.1" --vpc-id  --vswitch-id  \
  --project-spec 2C4G --storage-size 20 --pay-type Postpaid \
  --user-agent AlibabaCloud-Agent-Skills

Record: ProjectId (sbp-xxx), PublicConnectUrl, API Keys (store securely; do NOT print full API Keys in logs)

Timeout: Supabase project creation takes 5-10 minutes. Poll status until running:

> aliyun gpdb get-supabase-project --profile adbpg \
  --biz-region-id  --project-id  \
  --user-agent AlibabaCloud-Agent-Skills

Check Status field. Retry every 30 seconds until Status=running.

Step 2: Initialize Coaching Platform Database

Note: Steps 2-3 execute on Supabase Project, Steps 4-8 on ADBPG Instance. They are independent.

Modify whitelist, then connect via psql and execute schema from references/database-schema.md.

# Ask user for whitelist IP (do NOT use curl to external services)
# Example: "Please provide the IP address to add to the whitelist"
# Set whitelist
aliyun gpdb modify-supabase-project-security-ips --profile adbpg \
  --biz-region-id  --project-id  \
  --security-ip-list "" \
  --user-agent AlibabaCloud-Agent-Skills

Step 3: Insert Preset Coaching Domains

Execute SQL from references/database-schema.md via psql to insert coaching domains and coaching personas.

Step 4: Discover / Select / Create ADBPG Instance

4.1 Discover Existing Instances

aliyun gpdb describe-db-instances --profile adbpg \
  --biz-region-id  --page-size 100 \
  --user-agent AlibabaCloud-Agent-Skills

Filter results: DBInstanceStatus=Running AND VectorConfigurationStatus=enabled.

4.2 User Selects Instance

Present qualifying instances to user:

Available Instances (Running + Vector Enabled):
| # | Instance ID | Spec | Region | Status | Description |
|---|-------------|------|--------|--------|-------------|
| 1 | gp-xxxxx | 4C32G | cn-hangzhou | Running | Production |
| 2 | gp-yyyyy | 8C64G | cn-hangzhou | Running | Testing |

Select an instance, or enter "Create New".

• User selects existing → Go to Step 4.3
• User selects "Create New" → Go to Step 4.4
• No qualifying instances → Inform user, go to Step 4.4

4.3 Verify Selected Instance (when using existing)

aliyun gpdb describe-db-instance-attribute --profile adbpg \
  --db-instance-id  --region  \
  --user-agent AlibabaCloud-Agent-Skills

Confirm: DBInstanceStatus=Running + VectorConfigurationStatus=enabled. Then proceed to Step 5.

4.4 Create New Instance (when no existing or user chooses new)

Must present configuration and get user confirmation before execution:

💰 Cost note: Creating an instance incurs charges. Release or pause via ADBPG Console when not in use.

aliyun vpc describe-vswitches --profile adbpg \
  --biz-region-id  --zone-id  \
  --user-agent AlibabaCloud-Agent-Skills

Present VSwitch options to user, recommend the one with most available IPs.

After user confirms:

aliyun gpdb create-db-instance --profile adbpg \
  --biz-region-id  --zone-id  \
  --engine gpdb --engine-version "7.0" \
  --db-instance-mode StorageElastic --db-instance-category Basic \
  --instance-spec 4C16G --seg-node-num 2 \
  --storage-size 50 --seg-storage-type cloud_essd \
  --vpc-id  --vswitch-id  \
  --vector-configuration-status enabled --pay-type Postpaid \
  --user-agent AlibabaCloud-Agent-Skills

Timeout: Instance creation takes 10–15 minutes (max 30 min). Poll every 30–60 seconds:

> aliyun gpdb describe-db-instance-attribute --profile adbpg \
  --db-instance-id  --region  \
  --user-agent AlibabaCloud-Agent-Skills

Wait until DBInstanceStatus=Running.

Step 5: Configure Database Account

Check if the ADBPG instance already has a database account:

aliyun gpdb describe-accounts --profile adbpg \
  --db-instance-id  \
  --user-agent AlibabaCloud-Agent-Skills

Case A: No existing account → Create a new account:

Suggest account creation, confirm with user before executing:
- Account name: auto-generate ai_coaching_XX (XX = random 2-digit number), or user-specified
- Password: auto-generate a compliant password (8-32 chars, at least 3 character types, avoid !), or user-specified
- Example: Account: ai_coaching_01, Password: Coach3Acc#2x9K — Please confirm or provide your own.

⚠️ Important:
- Account name cannot be changed after creation — confirm carefully!
- Password can be reset via console, but save it securely now.
- This account will be used as ManagerAccount in Step 6.

aliyun gpdb create-account --profile adbpg \
  --db-instance-id  --region  \
  --account-name  --account-password '' \
  --user-agent AlibabaCloud-Agent-Skills

Case B: Account already exists → Inform the user. If the account was not created by the agent, ask the user for the existing account password before proceeding to Step 6.

Record: ManagerAccount and ManagerAccountPassword — these will be used in Step 6 for knowledge base initialization.

Step 6: Create Knowledge Base

Parameters to confirm for this step: Auto-generate the following, present to user for confirmation (user may modify), then execute.

| Parameter | Default | Notes |
|-----------|---------|-------|
| Namespace | ns_coaching | Namespace name, cannot be changed after creation |
| NamespacePassword | Auto-generate | Namespace password (same password rules); needed for uploads and coaching sessions |
| Collection | coaching_knowledge | Knowledge base name |
| EmbeddingModel | text-embedding-v4 | Embedding model |

Using the ManagerAccount and ManagerAccountPassword from Step 5, after user confirms the above parameters, execute:

# Initialize vector database
aliyun gpdb init-vector-database --profile adbpg \
  --biz-region-id  --db-instance-id  \
  --manager-account  --manager-account-password '' \
  --user-agent AlibabaCloud-Agent-Skills
# Create namespace
aliyun gpdb create-namespace --profile adbpg \
  --biz-region-id  --db-instance-id  \
  --manager-account  --manager-account-password '' \
  --namespace  --namespace-password '' \
  --user-agent AlibabaCloud-Agent-Skills
# Create document collection
aliyun gpdb create-document-collection --profile adbpg \
  --biz-region-id  --db-instance-id  \
  --manager-account  --manager-account-password '' \
  --namespace  --collection  \
  --embedding-model  --dimension 1024 \
  --user-agent AlibabaCloud-Agent-Skills

Step 7 (Optional): Upload Domain Knowledge Documents

If the user has domain knowledge documents (PDF/TXT/Markdown, etc.), upload them to the knowledge base to enhance coaching quality. This step can be skipped — proceed directly to Step 8 to start coaching.

aliyun gpdb upload-document-async --profile adbpg \
  --biz-region-id  --db-instance-id  \
  --namespace  --namespace-password '' \
  --collection  --file-name "domain_knowledge.pdf" \
  --file-url "https://example.com/knowledge.pdf" \
  --document-loader-name ADBPGLoader --chunk-size 500 --chunk-overlap 50 \
  --user-agent AlibabaCloud-Agent-Skills

Recommended documents by scenario: Sales methodologies, process guides (Workflow); Architecture patterns, design docs (Decision Support); Communication frameworks, best practices (Skill Development); Tech stack docs, onboarding guides (Onboarding).

Step 8: Start Coaching Session

Optional parameters for this step:

| Parameter | Default | Notes |
|-----------|---------|-------|
| Model | qwen-max | LLM model; use qwen-turbo for daily practice (lower cost) |
| TopK | 5 | RAG retrieval count |

Note: SourceCollection element MUST include Namespace field.

aliyun gpdb chat-with-knowledge-base --profile adbpg \
  --biz-region-id  --db-instance-id  \
  --model-params '{"Model": "", "Messages": [
    {"Role": "system", "Content": ""},
    {"Role": "user", "Content": ""}
  ]}' \
  --knowledge-params '{"SourceCollection": [{
    "Collection": "", "Namespace": "",
    "NamespacePassword": "", "QueryParams": {"TopK": }
  }]}' \
  --user-agent AlibabaCloud-Agent-Skills

Scenario Quick Reference

Success Verification

See references/verification-method.md for detailed verification steps.

Quick verification:

• Supabase project exists and is Running
• ADBPG instance has VectorConfigurationStatus=enabled
• Database tables exist (coaching_domains, coaching_personas, learners, coaching_sessions)
• Preset coaching domains are queryable
• ChatWithKnowledgeBase returns meaningful coaching responses

Best Practices

• Supabase for data, KB for AI — Session records through Supabase, coaching dialogue through RAG
• Coaching persona is key — Quality of system_prompt determines coaching effectiveness
• Always store session records — Write every coaching round for review and improvement
• All operations use --profile adbpg — Consistent credential management
• Team isolation with namespaces — Different teams use different Namespace
• TopK recommendation: 5 — Reduces token consumption
• Daily practice: qwen-turbo (low cost), assessments: qwen-max (high quality)
• Idempotent write operations — Before any resource creation (CreateSupabaseProject, CreateDBInstance, CreateAccount, CreateNamespace, etc.), always query first (Describe/List) to check if the resource already exists. Only create when the resource does not exist. This prevents duplicate resources on retry

References