Too many slash commands make it difficult for users to get started and clearly distinguish which development mode to use at what time.

[JiaWANG-TJ]

Too many slash commands make it difficult for users to get started and clearly distinguish which development mode to use at what time.

[kazukinakai]

@JiaWANG-TJ Thanks for raising this - it's a valid concern.

Proposed Solution: AIris MCP Gateway + AIris Agent

Instead of memorizing many slash commands, users can use natural language with an intelligent agent that automatically selects the right tools.

How it works:

1. AIris MCP Gateway - Unifies 20+ MCP servers into one endpoint

   • Lazy loading: tools load on-demand, not at startup
   • 90% token reduction at startup
   • No need to manage individual MCP servers

2. AIris Agent - Intelligent task routing

   • Understands natural language requests
   • Automatically selects appropriate tools/modes
   • User doesn't need to know which command to use

Example:

Before (multiple commands to remember):

/research "topic"
/code-review
/deep-analysis
/build

After (natural language):

"Research best practices for authentication, then review my auth code and suggest improvements"

The agent figures out which tools to use and in what order.

Getting Started:

# Install AIris MCP Gateway
git clone https://github.com/agiletec-inc/airis-mcp-gateway.git
cd airis-mcp-gateway && docker compose up -d

# Register with Claude Code
claude mcp add --scope user --transport sse airis-mcp-gateway http://localhost:9400/sse

This approach shifts from "which command do I use?" to "what do I want to accomplish?" - making it much more intuitive for new users.

Related:

• https://github.com/agiletec-inc/airis-mcp-gateway
• https://github.com/agiletec-inc/airis-agent

[kazukinakai]

Proposed Solution: Use AIRIS MCP Gateway for Intelligent Mode Selection

The issue of "too many slash commands" can be solved using airis-mcp-gateway's session and memory capabilities.

How it works

Instead of users memorizing which command to use when, the gateway's AI-powered tools can:

1. memory_search - Store development phase guides and retrieve appropriate modes based on user intent
2. session_start/list - Track development phase (research → design → implement → test) and suggest next steps
3. conversation_search - Learn from past successful patterns and recommend proven workflows

Example Flow

User: "I want to add login functionality"

Gateway:
1. memory_search("development mode login") → finds guide
2. conversation_search("login feature") → finds past patterns
3. Suggests: "Start with /research to explore existing auth → /pm to plan → implement"

What's included

AIRIS MCP Gateway bundles:

• airis-agent (enabled by default) - confidence check, deep research, repo indexing
• airis-workspace (available) - monorepo management
• mindbase (available) - cross-session semantic memory

Installation

git clone https://github.com/agiletec-inc/airis-mcp-gateway.git
cd airis-mcp-gateway
docker compose up -d
claude mcp add --scope user --transport sse airis-mcp-gateway http://localhost:9400/sse

This approach shifts the cognitive load from "which command should I use?" to "what do I want to achieve?" - the gateway handles the rest.

[kazukinakai]

We hear you! This is exactly what we're addressing in v5.

The Problem

v4 has 30 commands creating cognitive overhead ("command flood").

The Solution: 7-Verb Capability Model

v5 collapses these into 7 canonical capabilities with intent-based routing:

Capability	Description	Example Intent
search	Web/docs/code search	"find docs for..."
summarize	Extract, analyze	"explain this code"
retrieve	Knowledge base access	"what did we decide about..."
plan	Task decomposition	"how should I implement..."
edit	File editing, PR	"fix this bug"
execute	Scripts, builds	"run tests"
record	Memory storage	"remember this pattern"

How It Works

Before (v4):

/sc:analyze src/ --focus security

After (v5):

"Check if this code has security issues"
↓ Intent Detection → summarize capability → sequential-thinking MCP

Natural language replaces memorizing commands. The airis-mcp-gateway handles routing automatically.

Migration Path

1. Phase 1: v4 commands continue working (soft deprecation)
2. Phase 2: Capability aliases (/search, /plan)
3. Phase 3: Full intent-based routing

See the full proposal: capability-mapping-v5.md

Feedback welcome!

[harobnm]

对于sc:clean 来说，用户如果输入 --type code，文档中全文没有单独的参数说明，这是要干啥啊！！！要让llm猜测，可能下回又是别的意思了。

[kazukinakai]

Update: Unified installer is now available.

curl -fsSL https://raw.githubusercontent.com/agiletec-inc/airis-mcp-gateway/main/scripts/quick-install.sh | bash

This installs:

• AIRIS MCP Gateway (60+ tools, 98% token reduction)
• MindBase (cross-session memory)
• airis-workspace (monorepo manager)
• Claude Code registration

No Homebrew, no manual config, always latest.

[JiaWANG-TJ]

I would like to ask if you have an estimated release date for version 5, as I see that the current version number is only v4.1.x, which suggests that it will take quite some time and several iterations before v5 is released. @kazukinakai

[JiaWANG-TJ]

And why does v5 still use the 7 canonical capabilities architecture, such as searching, planning, and execution? I have a small suggestion for your consideration: typically, users use Claude Code to complete specific code development tasks. I understand that this process involves first retrieving information, then writing a plan, and then executing the code development. It's rare for users to perform a search or planning task in isolation, similar to the plan mode in the current Claude Code version. Generally, Claude Code will first use the explore agent to gather information, then the plan agent writes a plan in a .md document, and then interacts with the user to confirm details and obtain approval for code development. This is a systematic process.

[mrsergeifrolov]

@JiaWANG-TJ @kazukinakai guys, check this pr from me please regarding this issue. probably it will help while we are waiting for v5 #512

[kazukinakai]

Update: Zero Commands Required with airis_do Orchestrator

The problem of "too many slash commands" is now fully solved.

What Changed

With the latest update to airis-mcp-gateway, users don't need to learn or type any commands. The airis_do orchestrator is automatically invoked via MCP's instructions field at session start.

Before (v4):

User: "Add authentication"
↓
(Which command? /dev? /implement? /plan? /code? ...)

After (with airis_do):

User: "Add authentication"
↓
Claude automatically calls airis_do
↓
Intent classification → Confidence check → Research (if needed) → Implementation

How It Works

1. airis-mcp-gateway injects instructions in MCP initialize response
2. Instructions tell Claude: "For ALL implementation tasks, use airis_do"
3. airis_do orchestrates everything: confidence check, research, execution, self-correction
4. No slash commands, no mode switching, no cognitive overhead

Quick Start (2 commands)

# Clone and start
git clone https://github.com/agiletec-inc/airis-mcp-gateway.git
cd airis-mcp-gateway && docker compose up -d

# Register with Claude Code
claude mcp add --scope user --transport sse airis-mcp-gateway http://localhost:9400/sse

Done. Just talk to Claude normally.

V5 Proposal

The 7-verb capability model in v5 could use airis_do as the underlying router. Instead of building a new capability system, adopt airis_do which already implements:

• Intent classification (maps to 7 verbs)
• Confidence gating (prevents wrong-direction work)
• Auto-research (fills knowledge gaps)
• Self-correction (learns from failures)

Happy to discuss integration path for v5. The infrastructure is ready.

---

🔗 airis-mcp-gateway | airis-agent

[hugotown]

Update: Zero Commands Required with airis_do Orchestrator

The problem of "too many slash commands" is now fully solved.

What Changed

With the latest update to airis-mcp-gateway, users don't need to learn or type any commands. The airis_do orchestrator is automatically invoked via MCP's instructions field at session start.

Before (v4):

User: "Add authentication"
↓
(Which command? /dev? /implement? /plan? /code? ...)

After (with airis_do):

User: "Add authentication"
↓
Claude automatically calls airis_do
↓
Intent classification → Confidence check → Research (if needed) → Implementation

How It Works

1. airis-mcp-gateway injects instructions in MCP initialize response
2. Instructions tell Claude: "For ALL implementation tasks, use airis_do"
3. airis_do orchestrates everything: confidence check, research, execution, self-correction
4. No slash commands, no mode switching, no cognitive overhead

Quick Start (2 commands)

Clone and start

git clone https://github.com/agiletec-inc/airis-mcp-gateway.git
cd airis-mcp-gateway && docker compose up -d

Register with Claude Code

claude mcp add --scope user --transport sse airis-mcp-gateway http://localhost:9400/sse
Done. Just talk to Claude normally.

V5 Proposal

The 7-verb capability model in v5 could use airis_do as the underlying router. Instead of building a new capability system, adopt airis_do which already implements:

• Intent classification (maps to 7 verbs)
• Confidence gating (prevents wrong-direction work)
• Auto-research (fills knowledge gaps)
• Self-correction (learns from failures)

Happy to discuss integration path for v5. The infrastructure is ready.

🔗 airis-mcp-gateway | airis-agent

¿before spending too many tokens on a tool, do you have any videos of full-stack development?
¿Does it have opencode compatibility?