/project-architect

name: project-architect
description: >
  Documentation-first project planning that produces implementation-ready blueprints and
  single-shot coding agent prompts. Generates 4 interconnected docs — SPECIFICATION.md,
  IMPLEMENTATION.md, TASKS.md, BRANDING.md — plus a self-contained PROMPT.md for autonomous
  execution by any coding agent. Includes interactive tech stack selection, design pattern
  recommendations, and architecture decisions with trade-off analysis. Trigger when user
  wants to: plan a project, create specs, architect a system, break work into tasks, choose
  a tech stack, get design pattern advice, generate a coding agent prompt, or do
  documentation-first development. Phrases: "plan my project", "spec this out", "architect",
  "help me plan", "what stack should I use", "generate a prompt", "break this into tasks",
  "project docs", "I want to build X".
license: MIT
compatibility: >
  Works with any coding agent that supports file read/write and user interaction.
  Optimized for WrongStack and Claude Code, also compatible with Cursor, Gemini CLI, Codex,
  and other agentskills.io-compatible agents.
metadata:
  author: ersinkoc
  version: "1.1.0"
  category: development

Project Architect

A documentation-first project planning system that produces **implementation-ready blueprints** and **agent-ready prompts**. The philosophy: think deeply, document thoroughly, then execute with zero ambiguity. The final PROMPT.md is self-contained and portable — hand it to **WrongStack**, **Claude Code**, or any agent that reads and writes files.

Output Pipeline

[Discovery]  →  SPECIFICATION.md  →  IMPLEMENTATION.md  →  TASKS.md  →  BRANDING.md
                  (The What)           (The How)           (The Work)    (Identity)
                       ↓                    ↓                   ↓
                       └────────────────────┴───────────────────┘
                                          ↓
                                     PROMPT.md
                             (Single-Shot Agent Prompt)

Each document feeds the next. The final PROMPT.md synthesizes all documents into a single prompt optimized for single-shot execution by any coding agent (WrongStack, Claude Code, Cursor, and others).

Reference Files

Read the appropriate reference file before generating each document:

| Phase | Reference File | When to Read | |-------|---------------|--------------| | Discovery | `${CLAUDE_PLUGIN_ROOT}/references/elicitation-guide.md` | Before asking any questions | | Tech Stack | `${CLAUDE_PLUGIN_ROOT}/references/tech-stacks.md` | When user needs stack selection help | | Patterns | `${CLAUDE_PLUGIN_ROOT}/references/design-patterns.md` | When making architecture decisions | | Specification | `${CLAUDE_PLUGIN_ROOT}/references/specification-guide.md` | Before generating SPECIFICATION.md | | Implementation | `${CLAUDE_PLUGIN_ROOT}/references/implementation-guide.md` | Before generating IMPLEMENTATION.md | | Tasks | `${CLAUDE_PLUGIN_ROOT}/references/tasks-guide.md` | Before generating TASKS.md | | Branding | `${CLAUDE_PLUGIN_ROOT}/references/branding-guide.md` | Before generating BRANDING.md | | Prompt | `${CLAUDE_PLUGIN_ROOT}/references/agent-prompt.md` | Before generating PROMPT.md |

Workflow

Phase 0: Discovery & Elicitation

Read `${CLAUDE_PLUGIN_ROOT}/references/elicitation-guide.md` for the full question framework.

Before writing anything, understand the project through structured conversation. Use `AskUserQuestion` tool aggressively for choices — the user should tap, not type, whenever possible.

**Minimum understanding before any document generation:** 1. What does this project do? (elevator pitch) 2. Who is it for? (target audience) 3. What's the project type? (web app, CLI, library, API, mobile, desktop, infra tool) 4. What's the scope? (MVP vs full product) 5. Tech stack direction (or "help me choose")

**If the user says "help me choose a stack":** Read `${CLAUDE_PLUGIN_ROOT}/references/tech-stacks.md` and run the interactive stack selection flow. Present options with trade-offs using `AskUserQuestion`.

Phase 1: SPECIFICATION.md

Read `${CLAUDE_PLUGIN_ROOT}/references/specification-guide.md` before generating.

Defines **what** the project is. Technology-aware but not implementation-detailed. After generating, pause for user review.

Phase 2: IMPLEMENTATION.md

Read `${CLAUDE_PLUGIN_ROOT}/references/implementation-guide.md` AND `${CLAUDE_PLUGIN_ROOT}/references/design-patterns.md` before generating.

Translates specification into **how** to build it. This is where you:

• Recommend design patterns based on the project's needs (consult patterns reference)
• Define concrete directory structures with file-by-file purpose
• Choose dependencies with rationale
• Define module interfaces and data flows
• Include code snippets for critical patterns (signatures, types, structural examples)

**Pattern Recommendations:** Consult `${CLAUDE_PLUGIN_ROOT}/references/design-patterns.md` and recommend specific patterns with rationale. Don't just name-drop — show how each pattern applies to THIS project with a short code sketch.

Phase 3: TASKS.md

Read `${CLAUDE_PLUGIN_ROOT}/references/tasks-guide.md` before generating.

Converts implementation into **ordered work items**. Each task must be:

• Completable by a coding agent in a single session
• Self-contained with full context
• Ordered by strict dependency chain
• Include the exact files to create/modify

Phase 4: BRANDING.md (Optional)

Read `${CLAUDE_PLUGIN_ROOT}/references/branding-guide.md` before generating.

Only generate if user wants it or the project is user-facing.

Phase 5: PROMPT.md (Always Generate)

Read `${CLAUDE_PLUGIN_ROOT}/references/agent-prompt.md` before generating.

**This is the most critical output.** Synthesize all documents into a single-shot prompt that any coding agent (WrongStack, Claude Code, Cursor, …) can execute to build the entire project from scratch. The