Skip to main content
skill CX Codex

create-app · sethgammon — Claude Code skill: what it does, how to install

create-app — a skill for Claude Code and Codex that builds applications from a single prompt. It routes work across five distinct tiers, passing through PRD, architecture, and verified execution phases.

MIT ★ 937 updated 2026-11-05

Install into Claude Code · Plugin marketplace All methods and platforms ↓
/plugin marketplace add SethGammon/Citadel
/plugin install citadel@citadel-local
Open on GitHub

What create-app does

The skill operates in Claude Code and Codex, taking an app description and guiding it through PRD formulation, architectural planning, and execution via Archon. Depending on the input, it picks one of five tiers: blank scaffolding, guided milestones with user checkpoints, template loading from .planning/_templates/app-types/, fully autonomous builds, or in-place feature additions.

Autonomous execution activates circuit breakers: three failed attempts force a strategy shift, while five or more type errors pause the process entirely. When modifying existing codebases, Phase 0 records a baseline typecheck and test count, ensuring no new regressions pass into the final deliverable.

What it helps with

  • scaffold and build a full application from a plain-text prompt
  • add features to existing codebases without introducing regressions
  • produce structured PRD and architecture specs before coding begins
  • verify implementation against machine-verifiable acceptance criteria

Example prompts for the agent

  • /create-app build me a personal finance dashboard with auth
  • /create-app add payments integration to this project
  • /create-app help me create a real-time markdown editor

How to install create-app

1Plugin marketplace

/plugin marketplace add SethGammon/Citadel
/plugin install citadel@citadel-local

This installs the whole citadel plugin with all its skills

2Skills CLI (Vercel)

npx skills add SethGammon/Citadel --skill create-app -a claude-code

3Manually: put the files in place

~/.claude/skills/create-app/SKILL.md

Copy the whole skill folder; for one project use .claude/skills/create-app/

Step by step, with commands for each platform: How to install skills in Claude Code: five ways

Questions about create-app

How to install create-app in Claude Code?

Add the marketplace source with /plugin marketplace add SethGammon/Citadel, then install the plugin using /plugin install citadel@citadel-local.

What does create-app do?

It orchestrates end-to-end development by routing your prompt through PRD drafting, architecture design, and Archon task execution with step-by-step verification.

How does create-app handle existing projects?

It inspects existing files, records baseline test and typecheck results in Phase 0, and blocks phases if any new type errors or test failures appear.

Source text

SKILL.md · MIT
---
name: create-app
license: MIT
description: >-
  End-to-end app creation from a single description. Five tiers: blank project,
  guided, templated, fully generated, or feature addition to existing codebase.
  Routes through PRD, architecture, and Archon campaign with verification at every step.
user-invocable: true
auto-trigger: false
trigger_keywords:
  - create app
  - build app
  - build me
  - make an app
  - new app
  - generate app
  - add auth
  - add payments
  - integrate
last-updated: 2026-11-04
---

# /create-app — From Description to Verified Application

## Orientation

**Use when:** building a new app from scratch -- full scaffold, design system, and feature set from a single description.
**Don't use when:** adding a feature to an existing app (use /marshal or /archon); generating a single component (use /scaffold).

## Tier Detection

Use when the user wants to create an app, add a feature, or scaffold a project. Classify the input into one of five tiers:

### Tier 1: Blank Project
- Trigger: "create a blank project", "new project", "scaffold"
- Action: Run /scaffold with stack detection. No PRD, no architecture.

### Tier 2: Guided
- Trigger: "I want to build...", "help me create...", description with questions
- Action: /prd → user approves → /architect → user approves → /archon
- Human checkpoints: after PRD, after architecture, before each major phase.

### Tier 3: Templated
- Trigger: Describes a well-known app type ("a todo app", "a blog", "a dashboard with auth")
- Action: Load template PRD if available → /architect with template defaults → /archon
- Template detection: check `.planning/_templates/app-types/` for matching templates.
  If no template matches, fall through to Tier 2.

### Tier 4: Generated (Full Autonomy)
- Trigger: "build me [detailed description]", "create [app] and deploy it"
- Action: /prd (minimal questions) → /architect (auto-approve if confidence high) → /archon with self-correction loop
- Human checkpoints: after PRD only. Architecture and execution are autonomous.
- Safety: all Archon self-correction mechanisms active. Direction alignment every
  2 phases. Quality spot-checks every phase. Circuit breakers armed.

### Tier 5: Feature Addition (Existing Codebase)
- Trigger: existing project + feature description ("add auth", "add a dashboard", "add dark mode")
- Detection: project has source files (src/, app/, lib/, package.json with deps) AND description is a feature, not a standalone app
- Action: /prd in feature mode → /architect in existing codebase mode → /archon
- Key differences from greenfield tiers:
  - PRD reads existing codebase before asking questions
  - Architecture describes changes to existing files, not a standalone system
  - Phase 0 is always "Baseline" — record current typecheck/test state
  - Every phase end condition includes "no new typecheck errors" + "existing tests pass"
  - Risk register always includes "regression in existing functionality"
- Human checkpoints: after feature spec (PRD). Architecture can auto-approve if
  the feature is well-scoped and all conditions are machine-verifiable.

### Tier Classification

| Input Pattern | Tier |
|---|---|
| "blank project", "scaffold", "new empty" | 1 |
| "help me build", "I want to create", "guide me" | 2 |
| "todo app", "blog", "dashboard", well-known app type | 3 |
| "build me [detailed]", "create [app]", confident description | 4 |
| "add [feature]", "implement [feature]", existing project + feature description | 5 |
| Ambiguous | Default to Tier 2 (safest) |

## Protocol

### Step 1: CLASSIFY

Read the user's input. Determine the tier. Announce what you'll do in plain language — not tier numbers. If the classification is wrong, accept user overrides ("just scaffold it" → Tier 1, "just build it" → Tier 4).

### Proposal-only and pre-approval turns

Use the host session's configured effort level for this routing and proposal
turn. Downstream skills keep their own effort settings.

When the user requests a stack proposal or first steps before approving a build,
present that proposal after the initial project and matching-template check.
Choose reasonable defaults and state assumptions. Defer environment diagnostics,
additional stack/architecture research, dependency installation, and application
code until approval. A proposal-only turn ends at that approval boundary; it
does not enter the downstream build campaign.

### Step 2: EXECUTE TIER

**Tier 1:** Invoke /scaffold. Done.

**Tier 2:** /prd → user approves → /architect → user approves → Archon campaign → brief user after each major phase.

**Tier 3:** Check `.planning/_templates/app-types/` for template → present PRD, ask for changes → /architect with template defaults → Archon campaign.

**Tier 4:**
1. /prd express mode (0-1 questions)
2. User approves PRD (only mandatory checkpoint)
3. /architect (auto-approve if all end conditions are machine-verifiable)
4. Archon campaign with ALL safety systems active:
   - Direction alignment every 2 phases, quality spot-check every phase
   - Circuit breakers: 3 failures = new approach, 5+ type errors = park
5. Execute autonomously until complete or parked
6. Run full verification of all PRD end conditions, present results

**Tier 5 (Feature Addition):**
1. Read existing codebase — file tree, package.json, key entry points, existing patterns
2. /prd in feature mode (max 2 questions)
3. User approves feature spec (one mandatory checkpoint)
4. /architect in existing codebase mode — Phase 0 always "Baseline" (run typecheck + tests, record counts). Auto-approve if machine-verifiable.
5. Archon campaign — every phase end condition includes "no new typecheck errors vs baseline" and "existing tests pass"
6. On completion: verify all feature end conditions + baseline regression check, present results

### Step 3: VERIFY (All Tiers except 1)

Check each PRD end condition (run commands, check files, invoke /live-preview for visual checks). Report PASS / PARTIAL / FAIL with specifics.

### Step 4: DELIVER

Present: what was built, what was verified, what needs attention, how to run it, and suggested next step (e.g., /postmortem or deploy command).

## Quality Gates

- PRD exists and is approved before any code is written
- Architecture exists before Archon starts
- Every campaign phase has machine-verifiable end conditions
- Final verification checks all PRD end conditions
- User receives a clear report of what was built and what needs attention

## Fringe Cases

**Vague requirements**: Default to Tier 2 and ask clarifying questions before producing the PRD.

**Project already initialized**: Existing source files (src/, app/, package.json with deps) → automatically classify as Tier 5. Do not scaffold over an existing project.

**If .planning/ does not exist**: /prd and /architect will create it. If not possible, present inline and ask the user to run `/do setup` first.

**Tier misclassification**: Switch immediately on user correction without re-reading the input.

## Exit Protocol

After the campaign completes and verification runs, output:

```
---HANDOFF---
- App: {name}
- Built: {feature ledger summary}
- Verified: {N}/{total} end conditions passed
- Status: {complete | partial | failed}
- To run: {start command}
- Next: {suggested next step, e.g., /postmortem or deploy command}
- Reversibility: amber -- multi-tier creation, revert the creation commits
---
```

© SethGammon · MIT · SethGammon/Citadel

Similar skills

All skills for Agent setup: memory, context, orchestration →

create-app by other authors comes first.

create-app wondelai The create-app skill for Claude Code and five other platforms guides a new app from idea checks to an initial build plan. It asks for decisions along the way and records them in project files. CX Codex +1 2,600installs create-slack-app slackapi The create-slack-app Claude Code skill guides Slack app or agent setup with Bolt. It covers CLI checks, authentication, workspace selection, and project scaffolding. CX Codex +1 1,300installs create-code-app Microsoft create-code-app — a Microsoft skill that turns Claude Code into a Power Apps code app builder. Describe what you need and it scaffolds a React and Vite project, then deploys it to Power Platform. CX Codex +1 ★ 1,167repo stars project-docs coco-research project-docs scaffolds a standard 16-document folder ecosystem, sets up CLAUDE.local.md project memory, and audits repo structure against baseline documentation rules. CX Codex +1 ★ 328repo stars create-prd phuryn create-prd — a skill that generates a concise, decision-ready Product Requirements Document using an eight-section template: summary, contacts, background, objective, market segments, value propositions (JTBD-based), solution, and release. CX Codex +1 11,300installs ralph-tui-create-json subsy ralph-tui-create-json turns a PRD into prd.json for ralph-tui, with user stories, acceptance criteria, quality gates, and dependencies ready for execution. CX Codex +1 1,200installs project-creator itechmeat project-creator scaffolds structured technical documentation in a .project/ directory and guides you through specs, architecture, and user stories. CX Codex +1 113installs

From the same set55

The whole set →
merge-review merge-review scans the fleet worktree merge queue, flags files touched by more than one branch, and suggests a merge order before anything lands in main. CX Codex +1 ★ 937repo stars deploy-steward deploy-steward serializes PR merges into main, tracks CI gates, triggers deploys, and files repair tasks when checks fail. CX Codex +1 ★ 937repo stars prd prd turns an app idea into a structured product requirements document, asking about decisions that affect its shape. Use it before handing a project plan to Archon. CX Codex +1 ★ 937repo stars organize organize — a skill that audits and corrects project structure: checks file placement, moves files to match conventions, updates imports. Leaves code inside files untouched. CX Codex +1 ★ 937repo stars postmortem The postmortem skill reviews completed campaigns or incidents using project files, telemetry, and Git history. It shows what failed and which safeguards caught it. CX Codex +1 ★ 937repo stars map map generates a JSON index of a repository's files, exports, imports and dependency graph so an agent can locate relevant code without extra Glob or Grep calls. CX Codex +1 ★ 937repo stars phase-validator The phase-validator subagent for Claude Code compares phase handoff reports with exit conditions and returns a pass or fail verdict. It checks automated development pipelines. CX Codex ★ 937repo stars fleet fleet splits a big job into campaigns and runs them in waves of 2–3 agents, handing each wave's findings to the next. Fits when the tasks don't touch the same files. CX Codex ★ 937repo stars

Project sponsors

Companies that keep this analytics open