summonaikit
Back to Docs
// GETTING_STARTED

Getting Started

A two-pass setup for Claude Code, Cursor, and Codex, generate codebase skills, then wire your team workflow.

SummonAI Kit turns a generic AI coding agent into one that already knows your repo, your stack, and the way your team ships.

It works in two passes:

  1. The binary scans the codebase and writes a CLAUDE.md, a set of tech-stack skills, and specialized subagents.
  2. The slash command (/saikit-workflows) runs inside Claude Code, Cursor, or Codex and wires your team's actual operational stack, Linear, Stripe, Sentry, Slack, Cloudflare, and the rest, into hand-written skills, MCP servers, slash commands, and guardrail hooks.

By the end you've spent ten minutes and have an AI that opens a PR titled the way your team titles PRs, transitions tickets the way your team transitions tickets, and refuses to push to main.

//What Is SummonAI Kit?

A CLI plus a slash command. Together they generate every artifact Claude Code, Cursor, and Codex understand: instruction files (CLAUDE.md / .cursor/rules/ / AGENTS.md), skills, subagents, MCP wiring, slash commands, and hooks.

The CLI ships as a single signed binary, install once, run anywhere.

//Two Passes: Codebase, Then Workflow

The two passes are intentional. Codebase context (which framework, which ORM, how the monorepo is laid out) is a different problem from workflow context (which tracker, which branch convention, what fires after a PR merges). The first pass reads your code; the second pass interviews you about your team. Mixing them produces shallow output for both.

| Pass | Trigger | Source of truth | Output | | --- | --- | --- | --- | | 1, Codebase | summonaikit (binary) | Your repo | CLAUDE.md, <stack> skills, subagents | | 2, Workflow | /saikit-workflows (slash command) | Detection + your answers | <tool>-ops skills, .mcp.json, dev-cycle/SKILL.md, /start-task, /ship, /post-merge, hooks |

Pass 1 is required. Pass 2 is optional but is where the real leverage is.

//What It Generates

After both passes, your repo looks like this:

code
.
├── CLAUDE.md                          # Project overview, conventions, run commands
├── .mcp.json                          # MCP servers wired by the workflows pass
├── .claude/
│   ├── settings.json                  # Guardrail hooks (force-push block, .env block, ...)
│   ├── commands/
│   │   ├── saikit-workflows.md        # Re-run pass 2 anytime
│   │   ├── saikit-update.md           # Re-run pass 1 anytime
│   │   ├── start-task.md              # /start-task <TICKET>
│   │   ├── ship.md                    # /ship
│   │   └── post-merge.md              # /post-merge
│   ├── skills/
│   │   ├── react/SKILL.md             # Codebase skills (pass 1)
│   │   ├── nextjs/SKILL.md
│   │   ├── drizzle/SKILL.md
│   │   ├── stripe-ops/SKILL.md        # Ops skills (pass 2)
│   │   ├── linear-ops/SKILL.md
│   │   ├── sentry-ops/SKILL.md
│   │   └── dev-cycle/SKILL.md         # Synthesized playbook (pass 2)
│   └── agents/
│       ├── frontend-engineer.md
│       └── code-reviewer.md

The same artifacts can also write to .cursor/ and ~/.codex/ if you select Cursor or Codex as a target.

//Quick Start

First Pass, The Binary

code
# 1. Install Claude Code (or Cursor or Codex, at least one is required)
curl -fsSL https://claude.ai/install.sh | bash
claude auth login

# 2. Install SummonAI Kit
curl -fsSL https://cli.summonaikit.com/install.sh | bash

# 3. Run in your project
cd ~/code/your-app
summonaikit

The binary opens a browser to log in (device flow), detects which AI CLIs you have, asks which to install into, then analyzes the codebase. Five to ten minutes later you have a CLAUDE.md, per-stack skills, and subagents.

Second Pass, The Slash Command

Now open Claude Code, Cursor, or Codex inside the same project:

code
claude          # or: cursor . / codex

In the chat:

code
/saikit-workflows

The slash command:

  1. Asks which integration categories you care about, ticketing, payments, comms, error tracking, etc.
  2. Detects which tools in those categories are actually present in your repo (via deps, env vars, config files).
  3. Confirms with you, then asks two or three short questions per tool (e.g. for Stripe: "What's your test-vs-live policy? Which webhook events do you handle?").
  4. Installs each tool's ops skill (templated with your answers), wires its MCP server if available, and synthesizes a dev-cycle playbook from seven workflow questions (branch pattern, commit footer, PR title, ticket transitions, etc.).
  5. Optionally installs cross-tool commands (/start-task, /ship, /post-merge) and guardrail hooks.

You answer questions in chat. The slash command is a thin protocol executor, all the logic lives in the binary. Re-run it any time to add tools.

//What's New in 3.0.0

3.0.0 is the first release where the workflow pass exists. The numbers:

  • 110 supported tools across 14 categories
  • 33 hand-written ops skills (Stripe, Sentry, Linear, GitHub Issues, Slack, Cloudflare, Vercel, Supabase, Neon, Postgres, Drizzle, Prisma, Better Auth, Clerk, Resend, PostHog, Datadog, Polar, Notion, Confluence, Figma, HubSpot, Customer.io, Meta Ads, Intercom, Cal.com, Statsig, GSC, Profound, structured data, in-repo ADRs, …)
  • 47 official MCPs auto-detected, 23 of which are zero-config paste-URL HTTP endpoints
  • 7-question dev-cycle playbook synthesized into a project-level skill

See the Workflows Overview for the full picture.

//Next Steps

Getting Started | SummonAI Kit Docs