stack-architecture
SKILL.md
Stack Architecture Design
Design transport-agnostic handler systems with proper Result types and error taxonomy.
Process
Step 1: Understand Requirements
Gather information about:
- Transport surfaces — CLI, MCP, HTTP, or all?
- Domain operations — What actions does the system perform?
- Failure modes — What can go wrong? (maps to error taxonomy)
- External dependencies — APIs, databases, file system?
Step 2: Design Handler Layer
For each domain operation:
- Define input type (Zod schema)
- Define output type
- Identify possible error types (from taxonomy)
- Write handler signature:
Handler<Input, Output, Error1 | Error2>
Example:
// Input schema
const CreateUserInputSchema = z.object({
email: z.string().email(),
name: z.string().min(1),
});
// Output type
interface User {
id: string;
email: string;
name: string;
}
// Handler signature
const createUser: Handler<unknown, User, ValidationError | ConflictError>;
Step 3: Map Errors to Taxonomy
Map domain errors to the 10 categories:
| Domain Error | Stack Category | Error Class |
|---|---|---|
| Not found | not_found |
NotFoundError |
| Invalid input | validation |
ValidationError |
| Already exists | conflict |
ConflictError |
| No permission | permission |
PermissionError |
| Auth required | auth |
AuthError |
| Timed out | timeout |
TimeoutError |
| Connection failed | network |
NetworkError |
| Limit exceeded | rate_limit |
RateLimitError |
| Bug/unexpected | internal |
InternalError |
| User cancelled | cancelled |
CancelledError |
Step 4: Choose Packages
Packages are organized into three tiers:
Package Tiers
┌─────────────────────────────────────────────────────────────────┐
│ TOOLING TIER │
│ Build-time, dev-time, test-time packages │
│ @outfitter/testing │
└─────────────────────────────────────────────────────────────────┘
▲
│ depends on
┌─────────────────────────────────────────────────────────────────┐
│ RUNTIME TIER │
│ Application-specific packages for different deployment targets │
│ @outfitter/cli @outfitter/mcp @outfitter/daemon │
│ @outfitter/config @outfitter/logging @outfitter/file-ops │
│ @outfitter/state │
└─────────────────────────────────────────────────────────────────┘
▲
│ depends on
┌─────────────────────────────────────────────────────────────────┐
│ FOUNDATION TIER │
│ Zero-runtime-dependency core packages │
│ @outfitter/contracts @outfitter/types │
└─────────────────────────────────────────────────────────────────┘
| Tier | Packages | Dependency Rule |
|---|---|---|
| Foundation | contracts, types |
No @outfitter/* deps |
| Runtime | cli, mcp, daemon, config, logging, file-ops, state |
May depend on Foundation |
| Tooling | testing |
May depend on Foundation + Runtime |
Package Selection
| Package | Purpose | When to Use |
|---|---|---|
@outfitter/contracts |
Result types, errors, Handler contract | Always (foundation) |
@outfitter/types |
Type utilities, collection helpers | Type manipulation |
@outfitter/cli |
CLI commands, output modes, formatting | CLI applications |
@outfitter/mcp |
MCP server, tool registration | AI agent tools |
@outfitter/config |
XDG paths, config loading | Configuration needed |
@outfitter/logging |
Structured logging, redaction | Logging needed |
@outfitter/daemon |
Background services, IPC | Long-running services |
@outfitter/file-ops |
Secure paths, atomic writes, locking | File operations |
@outfitter/state |
Pagination, cursor state | Paginated data |
@outfitter/testing |
Test harnesses, fixtures | Testing |
Selection criteria:
- All projects need
@outfitter/contracts(foundation) - CLI applications add
@outfitter/cli(includes UI components) - MCP servers add
@outfitter/mcp - File operations need both
@outfitter/config(paths) and@outfitter/file-ops(safety)
Step 5: Design Context Flow
Determine:
- Entry points — Where is context created? (CLI main, MCP server, HTTP handler)
- Context contents — Logger, config, signal, workspaceRoot
- Tracing — How requestId flows through operations
Output Templates
Architecture Overview
Project: {PROJECT_NAME}
Transport Surfaces: {CLI | MCP | HTTP | ...}
Directory Structure:
├── src/
│ ├── handlers/ # Transport-agnostic business logic
│ │ ├── {handler-1}.ts
│ │ └── {handler-2}.ts
│ ├── commands/ # CLI adapter (if CLI)
│ ├── tools/ # MCP adapter (if MCP)
│ └── index.ts # Entry point
└── tests/
└── handlers/ # Handler tests
Dependencies:
├── @outfitter/contracts # Foundation (always)
├── @outfitter/{package-2} # {reason}
└── @outfitter/{package-3} # {reason}
Handler Inventory
| Handler | Input | Output | Errors | Description |
|---|---|---|---|---|
getUser |
GetUserInput |
User |
NotFoundError |
Fetch user by ID |
createUser |
CreateUserInput |
User |
ValidationError, ConflictError |
Create new user |
deleteUser |
DeleteUserInput |
void |
NotFoundError, PermissionError |
Remove user |
Error Strategy
Domain Errors → Stack Taxonomy:
{domain-error-1} → {stack-category} ({ErrorClass})
- When: {condition}
- Exit code: {code}
{domain-error-2} → {stack-category} ({ErrorClass})
- When: {condition}
- Exit code: {code}
Implementation Order
- Foundation — Install packages, create types
- Core handlers — Implement business logic with tests
- Transport adapters — Wire up CLI/MCP/HTTP
- Testing — Integration tests across transports
Constraints
Always:
- Recommend Result types over exceptions
- Map domain errors to taxonomy categories
- Design handlers as pure functions (input, context) → Result
- Consider all transport surfaces upfront
- Include error types in handler signatures
Never:
- Suggest throwing exceptions
- Design transport-specific logic in handlers
- Recommend hardcoded paths
- Skip error type planning
- Couple handlers to specific transports
Related Skills
outfitter-stack:stack-patterns— Reference for all patternsoutfitter:tdd— TDD implementation methodologyoutfitter-stack:stack-templates— Templates for components
Weekly Installs
6
Repository
outfitter-dev/agentsGitHub Stars
24
First Seen
Jan 29, 2026
Security Audits
Installed on
github-copilot6
opencode5
gemini-cli5
codex5
kimi-cli5
amp5