think-cog（认为齿轮）

Think Cog - Your Alfred for Complex Ideas

Complex ideas need iteration, not conversation. Think → Do → Review → Repeat.

CellCog is the thinking partner that completes the full loop — reasons with you, then executes across every modality: research, documents, visuals, data, prototypes. Review real output, refine your thinking, iterate on substance. Like having an Alfred who doesn't just advise — he builds.

For problems where you don't know the answer upfront and the solution emerges through doing, not just discussing.

How to Use

For your first CellCog task in a session, read the cellcog skill for the full SDK reference — file handling, chat modes, timeouts, and more.

OpenClaw (fire-and-forget):

result = client.create_chat(
    prompt="[your task prompt]",
    notify_session_key="agent:main:main",
    task_label="my-task",
    chat_mode="agent",
)

All agents except OpenClaw (blocks until done):

from cellcog import CellCogClient
client = CellCogClient(agent_provider="openclaw|cursor|claude-code|codex|...")
result = client.create_chat(
    prompt="[your task prompt]",
    task_label="my-task",
    chat_mode="agent",
)
print(result["message"])

Why Think-Cog Exists

Most CellCog skills follow a pattern: you know what you want → CellCog delivers it.

But some problems don't work that way:

• You have an idea but don't know if it's possible
• You're stuck and need a different perspective
• The solution emerges through exploration, not execution
• You need to reason through tradeoffs before committing

Think-cog is your worker agent for intellectual exploration. You're the manager agent providing direction. CellCog thinks and works alongside you.

How It Works

Unlike single-shot tasks, think-cog is conversational by design:

# Start the thinking session
result = client.create_chat(
    prompt="[your thinking prompt]",
    task_label="my-task",
    chat_mode="agent",
)

# Continue the conversation — each exchange deepens understanding
result = client.send_message(
    chat_id=result["chat_id"],
    message="What if we approached it from this angle instead?",
)

Think-cog expects back-and-forth conversation, not single requests. Each exchange deepens understanding.

See https://cellcog.ai for complete SDK API reference.

When to Use Think-Cog

Architecture & Technical Decisions

When you're weighing tradeoffs:

"I'm building a notification system that needs to handle 10M daily users. I'm torn between: 1. WebSocket connections for all users 2. Server-sent events with polling fallback 3. Push notifications only Help me think through the tradeoffs. My team is small (3 engineers) and we're on AWS."

Business Strategy

When the path forward isn't clear:

"My SaaS is growing but I'm not sure whether to: - Focus on enterprise sales (fewer customers, bigger deals) - Double down on self-serve (more customers, smaller ARPU) Current metrics: 500 customers, $50 ARPU, 2 enterprise deals in pipeline worth $50K each. Let's think through this together."

Creative Direction

When you're exploring possibilities:

"I want to create a video campaign for my coffee brand but I'm not sure what angle to take. The brand is: - Specialty single-origin coffee - Direct trade with farmers - Premium pricing ($25/bag) Target audience is 25-40 professionals who care about quality. Help me brainstorm directions before we commit to production."

Problem Debugging

When you're stuck:

"My machine learning model keeps overfitting despite: - Dropout layers - Data augmentation - Early stopping Here's my architecture: [details] Let's debug this together. What am I missing?"

Decision Making

When you need structured thinking:

"I have three job offers and I'm paralyzed by the decision: - Startup (lower pay, more equity, more risk) - Big tech (great pay, slower growth, stable) - Mid-stage scaleup (balanced, interesting problem) Help me build a framework to think through this."

The Think-Cog Philosophy

"It's hard to know what's actually possible before you try it."

This is CellCog's core insight. Complex problems rarely have obvious solutions. The best work happens when two smart entities—you and CellCog—exchange ideas, challenge assumptions, and discover answers together.

Think of it as pair programming for thinking:

• You provide direction and domain expertise
• CellCog provides reasoning depth and fresh perspectives
• Solutions emerge from the conversation

Chat Mode for Thinking

Use chat_mode="agent" for think-cog sessions.

Why not agent team? Agent team is designed for autonomous deep work with multiple internal passes. Think-cog is about collaborative iteration WITH you. Agent mode keeps the conversation flowing naturally.

The magic happens in the back-and-forth, not in autonomous processing.

Tips for Better Thinking Sessions

1. Share context generously: The more CellCog understands your situation, the better it can think alongside you.

1. State your constraints: Budget, timeline, team size, technical limitations—these shape what's possible.

1. Be honest about uncertainty: "I'm not sure if X is even feasible" is a great starting point.

1. Challenge the responses: Push back, ask "what about...", explore alternatives. Thinking improves through friction.

1. Let solutions emerge: Don't expect the answer in the first exchange. The best insights often come in message 4 or 5.

1. Capture decisions: When you reach clarity, summarize the conclusions. You can then use other CellCog skills to execute.

Think-Cog vs Other Skills

Think-cog is for when you're in discovery mode. Other skills are for when you're in execution mode.

Example Thinking Session

You:

"I'm building a two-sided marketplace for freelance designers. I'm stuck on the chicken-and-egg problem—how do I get designers without clients, and clients without designers?"

CellCog:

"This is the classic marketplace cold-start problem. Let me think through some approaches with you... Option 1: Supply-first (designers) - Pros: Designers are easier to acquire, content creates SEO value - Cons: They'll leave if no jobs come Option 2: Demand-first (clients) - Pros: Revenue from day one, can manually fulfill initially - Cons: Harder to acquire, need to deliver quality Option 3: Constrain the market - Focus on ONE design type (e.g., logo design only) - Build liquidity in a narrow niche before expanding What's your current thinking? And what resources do you have to invest in early growth?"

You:

"I have about $20K for initial marketing and can do some manual matching myself. The constrained market approach is interesting—I was thinking of going broad."

CellCog:

"That $20K + your time is actually a meaningful advantage. Let me build on the constrained market idea... What if you started with: - Niche: Startup logo design only - Supply: Recruit 20-30 designers who specialize in startup branding (curated, not open signup) - Demand: Target YC companies, indie hackers, ProductHunt launchers The manual matching you'd do becomes a feature, not a bug—'curated matches by founders who understand startups.' This gives you: - Clear positioning - Easier SEO/content ('startup logo design' vs 'design marketplace') - Higher willingness to pay from funded startups - Story for press/ProductHunt Thoughts? Or should we explore other niches?"

The conversation continues, ideas refine, and eventually you arrive at a clear strategy to execute.

If CellCog is not installed

Run /cellcog-setup (or /cellcog:cellcog-setup depending on your tool) to install and authenticate. OpenClaw users: Run clawhub install cellcog instead. Manual setup: pip install -U cellcog and set CELLCOG_API_KEY. See the cellcog skill for SDK reference. Other agents: pip install cellcog and set CELLCOG_API_KEY. See https://cellcog.ai for details.