home / skills / willsigmon / sigstack / sigstack-quickstart

This skill accelerates Claude-based workflows by teaching rapid session loops, data fetching, and parallel agent spawning for faster, reproducible results.

npx playbooks add skill willsigmon/sigstack --skill sigstack-quickstart

Review the files below or copy the command above to add this skill to your agents.

Files (1)
SKILL.md
3.6 KB
---
name: Sigstack Quickstart
description: Get started with sigstack in 5 minutes
allowed-tools: Read, Bash
model: haiku
---

# Sigstack Quickstart

**From zero to productive in 5 minutes.**

## What is Sigstack?

The vibe coder's operating system for Claude.

```
You describe → Claude builds → Screenshots verify → Repeat
```

No traditional coding knowledge needed.
5000+ hours of Claude experience? You're ready.

## The 3 Things to Remember

### 1. Screenshot Everything
```
After any change:
1. Screenshot it
2. "What's wrong?"
3. Fix what Claude finds
4. Repeat

This is THE workflow.
```

### 2. Let MCP Fetch Data
```
Don't copy-paste from:
- Databases (use Supabase MCP)
- GitHub (use GitHub MCP)
- Docs (use WebFetch)

Claude fetches directly. Fewer tokens.
```

### 3. Spawn Agents for Parallel Work
```
Big search? Spawn 5-10 agents.
PR review? Spawn specialists.
Feature build? Spawn workers.

One message, multiple Task calls.
```

## Your First Session

### 1. Start Clean
```
"What's the state of [project]?"

Claude checks git, files, context.
```

### 2. Set Focus
```
"Today: [specific thing]"

Narrow context = better results.
```

### 3. Work the Loop
```
"Build [thing]"
[screenshot]
"What's wrong?"
[fix]
[screenshot]
"Better?"
```

### 4. End Clean
```
"Checkpoint"

Saves state for next session.
```

## Essential Commands

| Command | What It Does |
|---------|--------------|
| `Status` | Check current state |
| `Checkpoint` | Save session |
| `Resume` | Continue from checkpoint |
| `Focus: [area]` | Narrow context |
| `Remember: [thing]` | Store in memory |
| `Recall [topic]` | Retrieve from memory |

## Model Selection

```
Simple task → Haiku (fast, cheap)
Code work → Sonnet (default)
Architecture → Opus (deep thinking)
```

Skills specify this in frontmatter.
You don't need to choose manually.

## Your MCP Servers

Check what's available:
```
Supabase → Database
GitHub → Repos, PRs
Memory → Persistent knowledge
Playwright → Browser automation
```

Use them. Don't copy-paste data.

## Your Voice Stack

```
Typeless → Daily dictation
Sled → Mobile voice to Claude Code
```

Speak your ideas. Code appears.

## Skill Invocation

```
/vibe-coder-qa       → Default QA workflow
/token-optimizer     → Reduce costs
/agent-patterns      → When to spawn
/session-handoff     → End sessions properly
```

Skills are reusable prompts.
Create your own when patterns repeat.

## Common Mistakes

### ❌ Don't
```
- Explain everything every session
- Copy-paste large data
- Work sequentially when parallel works
- Skip the screenshot step
- Close without checkpointing
```

### ✓ Do
```
- Use CLAUDE.md for project rules
- Use MCP for external data
- Spawn agents for parallel work
- Screenshot after every change
- Checkpoint before closing
```

## Quick Wins

### Save 90% Tokens
```
Before: "Here's my 5000-line database export..."
After: "Query Supabase for users created today"
```

### 5x Faster Search
```
Before: Search one module at a time
After: Spawn 5 agents, one per module
```

### Never Re-Explain
```
Before: "As I mentioned, we use SwiftUI..."
After: [It's in CLAUDE.md, auto-loaded]
```

## Next Steps

1. Read `/Users/wsig/Developer/sigstack/SIGSTACK.md`
2. Try the Vision QA workflow on your next change
3. Use "Checkpoint" before closing
4. Create a skill when you repeat something 3+ times

## The Sigstack Promise

```
Describe the outcome.
Claude handles implementation.
Screenshots verify results.
Agents parallelize work.
Memory preserves knowledge.
Voice captures ideas.

Build faster. Ship more. Learn always.
```

Use when: Getting started, onboarding others, refreshing workflow

Overview

This skill gets you productive with Sigstack in five minutes by guiding a repeatable Claude-driven workflow. It focuses on fast iteration with screenshots, using MCP connectors for data, and spawning agents to parallelize work. The goal is faster development, lower token costs, and consistent sessions you can resume.

How this skill works

You describe the desired outcome and Claude performs implementation steps while you verify with screenshots. MCP connectors (Supabase, GitHub, WebFetch, Playwright) let Claude fetch data directly instead of pasting large payloads. Spawn multiple task agents for parallel searches or specialist reviews and use checkpoints to save session state for later resumption.

When to use it

  • Starting a new project session or onboarding to an existing repo
  • Working on code or architecture tasks that benefit from parallel agents
  • When you want to reduce token usage by letting MCP fetch external data
  • During iterative feature development where visual verification matters
  • Before closing work — to create a checkpoint and preserve context

Best practices

  • Screenshot after every meaningful change and ask “What’s wrong?”
  • Use MCP connectors (Supabase, GitHub, WebFetch) instead of copy-pasting data
  • Spawn multiple agents for big searches or simultaneous code reviews
  • Set a narrow daily focus to limit context and improve results
  • Checkpoint at the end of a session so state can be resumed later

Example use cases

  • Iterative UI fixes: make a change, screenshot, ask for problems, and repeat
  • PR reviews: spawn specialist agents to analyze tests, style, and security in parallel
  • Large-repo searches: run 5–10 agents, one per module, to find references quickly
  • Data-driven tasks: ask Supabase MCP for a query result instead of pasting exports
  • Onboarding: run a status check, set focus, and follow the work loop to bring new collaborators up to speed

FAQ

Do I need to know coding to use Sigstack?

No. The workflow is built for describing outcomes to Claude. Experience with Claude or familiarity with the project rules helps, but you don’t need to be a coder.

How do I avoid high token costs?

Let MCP connectors fetch data (Supabase, GitHub, WebFetch) rather than pasting large files. Use smaller models like Haiku for trivial tasks and reserve Sonnet/Opus for heavier work.