An Agent Skill for documentation-first project planning. Transforms a project idea into implementation-ready blueprints and a single-shot coding agent prompt.
Compatible with 40+ coding agents including WrongStack, Claude Code, Cursor, Gemini CLI, GitHub Copilot, Codex, and more.
WrongStack is a free, open-source AI coding agent with a Brain, a Memory, and a full toolbox. It reads your code, edits files, runs commands, and coordinates specialist agents — across six surfaces, from a plain terminal REPL to a cross-machine HQ dashboard. No subscription required, and you keep your hand on every permission.
curl -fsSL https://wrongstack.com/install.sh | sh # macOS / Linux — self-contained binaryirm https://wrongstack.com/install.ps1 | iex # Windows (PowerShell) — no Node.js needed| What you get | |
|---|---|
| 🌐 200+ LLM providers | Catalog pulled live from models.dev — Anthropic, OpenAI, Google, plus OAuth sign-in for Claude Pro/Max, ChatGPT and Copilot, and any OpenAI-compatible endpoint (Ollama, vLLM, LM Studio) |
| 🛠️ 67 built-in tools | Edits, lint/typecheck/test, execution, git, web, browser/E2E and a SQLite codebase index — every call gated by per-tool permissions |
| 🧠 SAGE memory | Project-wide long-term memory in SQLite/FTS5, anchored to files, symbols and commits — it gets better at your codebase over time |
| 🖥️ Six surfaces | Readline REPL · Ink/React TUI (--tui) · WebUI · SimpleUI · Desktop · cross-machine HQ (--hq) |
| 🤖 Fleet orchestration | A Director fans out specialist subagents over a project mailbox; eternal & parallel goal loops run until the contract verifies |
| 🔍 Chimera & Kanban | Auto-review agents that critique your diffs with severity-ranked findings, plus durable Kanban boards with atomic verification |
| 🔐 Secure by default | Encrypted secrets at rest, project-root containment, opt-in YOLO mode — MIT licensed, TypeScript-strict |
📊 The perfect pairing: This band tells you exactly where your Claude limits stand — and WrongStack keeps you moving when they close in. It reads plan windows for Claude, ChatGPT, Copilot, Z.AI and more right in its own statusline and quota page, and when one provider runs dry, fallback chains rotate you onto the next model automatically. Watch the band, dodge the wall.
Given a project idea, Project Architect walks you through an interactive discovery process and generates 5 interconnected documents:
[Discovery] -> SPECIFICATION.md -> IMPLEMENTATION.md -> TASKS.md -> BRANDING.md
(The What) (The How) (The Work) (Identity)
| | |
+-------------------+------------------+
|
PROMPT.md
(Single-Shot Agent Prompt)
| Document | Purpose |
|---|---|
| SPECIFICATION.md | What the project is, features, data model, API surface |
| IMPLEMENTATION.md | Tech stack, design patterns, directory structure, schemas |
| TASKS.md | Ordered work items, each completable in a single agent session |
| BRANDING.md | Name, colors, typography, voice (optional, for user-facing projects) |
| PROMPT.md | Self-contained prompt to build the entire project from scratch |
- Interactive tech stack advisor -- presents options with trade-offs, lets you choose
- Design pattern recommendations -- matched to your project's specific needs with code sketches
- Scales to project size -- weekend hack gets 15 tasks, enterprise system gets 100+
- Agent-optimized output -- every task lists exact files to create/modify with acceptance criteria
- Pause-and-review flow -- generates each document, waits for approval before continuing
The easiest way to install, works with all agentskills.io-compatible agents:
npx skills add ersinkoc/project-architectThis auto-detects your coding agent and installs to the correct location.
Install for a specific agent:
npx skills add ersinkoc/project-architect --agent wrongstack
npx skills add ersinkoc/project-architect --agent claude-code
npx skills add ersinkoc/project-architect --agent cursor
npx skills add ersinkoc/project-architect --agent codexInstall globally (all projects):
npx skills add ersinkoc/project-architect --globalRepository: github.com/ersinkoc/project-architect
Add to your project's .claude/settings.json:
{
"plugins": [
"/absolute/path/to/project-architect"
]
}Or add to ~/.claude/settings.json for global availability.
Clone the repo and copy/symlink the skill directory into your agent's skills folder:
| Agent | Skills Directory |
|---|---|
| WrongStack | .wrongstack/skills/ or ~/.wrongstack/skills/ |
| Claude Code | .claude/skills/ or ~/.claude/skills/ |
| Cursor | .cursor/skills/ or ~/.cursor/skills/ |
| GitHub Copilot | .github/skills/ |
| Codex | .codex/skills/ |
| Gemini CLI | .gemini/skills/ |
| OpenCode | .opencode/skills/ |
| Generic | .agents/skills/ or ~/.agents/skills/ |
# Example: install for WrongStack globally
git clone https://github.com/ersinkoc/project-architect.git
ln -s $(pwd)/project-architect ~/.wrongstack/skills/project-architect
# Example: install for Claude Code globally
ln -s $(pwd)/project-architect ~/.claude/skills/project-architect
# Example: install for Cursor in a project
ln -s $(pwd)/project-architect .cursor/skills/project-architectStart a conversation with your coding agent and describe what you want to build:
> plan my project: a CLI tool for managing dotfiles across machines
Or use any of these trigger phrases:
- "plan my project"
- "spec this out"
- "architect a system for..."
- "help me plan"
- "what stack should I use"
- "generate a prompt"
- "break this into tasks"
- "I want to build X"
- Discovery -- Agent asks structured questions about your project (type, scope, stack, features)
- SPECIFICATION.md -- Generated and presented for review
- IMPLEMENTATION.md -- Tech decisions, patterns, directory structure
- TASKS.md -- Ordered work breakdown with file lists and acceptance criteria
- BRANDING.md -- Optional identity guide (colors, typography, voice)
- PROMPT.md -- Everything synthesized into a single executable prompt
You review and approve each document before the next is generated.
You don't have to run the full pipeline:
| What You Say | What Happens |
|---|---|
| "Just the spec" | Generates SPECIFICATION.md only |
| "Skip to tasks" | Lightweight spec + impl, then detailed tasks |
| "Just give me a prompt" | Condensed discovery, straight to PROMPT.md |
| "Help me choose a stack" | Interactive tech stack selection only |
| "What patterns should I use?" | Design pattern consultation |
The skill includes detailed guides that inform each generation phase:
| File | Purpose |
|---|---|
references/elicitation-guide.md |
Question framework for project discovery |
references/tech-stacks.md |
Interactive tech stack selection with trade-offs |
references/design-patterns.md |
Pattern catalog with selection guide |
references/specification-guide.md |
Template and rules for SPECIFICATION.md |
references/implementation-guide.md |
Template and rules for IMPLEMENTATION.md |
references/tasks-guide.md |
Template and rules for TASKS.md |
references/branding-guide.md |
Template and rules for BRANDING.md |
references/agent-prompt.md |
Template and rules for PROMPT.md |
project-architect/
├── SKILL.md # Skill definition (agentskills.io format)
├── plugin.json # Claude Code plugin manifest
├── LICENSE # MIT License
├── README.md # This file
└── references/
├── elicitation-guide.md # Discovery question framework
├── tech-stacks.md # Tech stack advisor
├── design-patterns.md # Pattern catalog
├── specification-guide.md # Spec template
├── implementation-guide.md # Implementation template
├── tasks-guide.md # Tasks template
├── branding-guide.md # Branding template
└── agent-prompt.md # Prompt template
This skill follows the Agent Skills specification and is compatible with:
- WrongStack (terminal-native AI coding agent — wrongstack.com)
- Claude Code (plugin + skill)
- Cursor
- GitHub Copilot
- OpenAI Codex
- Gemini CLI
- OpenCode
- Kiro
- And 30+ more agents via agentskills.io
MIT