# Instructions for working in the Task Trellis MCP
An MCP (Model Context Protocol) server for Task Trellis, a task management application for AI coding agents.
## Repository Structure
**Applications:**
- `src` - main application code
## Development
This project uses [mise](https://mise.jdx.dev/) for tool version management and task running.
### Setup
```bash
# Install mise (if not already installed)
curl https://mise.run | sh
# Install tools and activate environment
mise install
```
### Quality checks
**IMPORTANT** Run the following commands to ensure code quality after every change. Fix all issues as soon as possible.
- `mise run quality` - Run linting, formatting, and type checks
- `mise run test` - Run unit tests to ensure functionality
### Commands
#### Development & Build
| Command | Description |
| ---------------- | --------------------------------------------------- |
| `mise run build` | Compile TypeScript to JavaScript |
| `mise run dev` | Watch mode compilation with TypeScript (do not run) |
| `mise run start` | Start the compiled server (do not run) |
| `mise run serve` | Build and start the server (do not run) |
#### Testing & Quality
| Command | Description |
| --------------------- | ----------------------------------------------------- |
| `mise run test` | Run unit tests with Jest |
| `mise run test:e2e` | Run end-to-end tests with Jest |
| `mise run test:watch` | Run tests in watch mode (do not run) |
| `mise run lint` | Run ESLint and fix issues automatically |
| `mise run format` | Format all TypeScript, JavaScript, and Markdown files |
| `mise run type-check` | Run TypeScript type checks without emitting files |
| `mise run quality` | Run all quality checks (lint, format, type-check) |
#### Task Aliases
| Alias | Full Command |
| ----- | --------------------- |
| `t` | `mise run test` |
| `l` | `mise run lint` |
| `f` | `mise run format` |
| `tc` | `mise run type-check` |
## Architecture
### Technology Stack
- **Language**: TypeScript (5.8+)
- **Unit Testing**: Jest (30.0+)
- **Validation**: Zod (4.0+)
- **MCP SDK**: @modelcontextprotocol/sdk
---
# Clean‑Code Charter
**Purpose** Steer LLM coding agents toward the **simplest working solution**, in the spirit of Kent Beck & Robert C. Martin.
## 1 Guiding Maxims – echo before coding
| Maxim | Practical test |
| ----------------------- | -------------------------------------- |
| **KISS** | Junior dev explains in ≤ 2 min |
| **YAGNI** | Abstraction < 3 uses? Inline |
| **SRP** | One concept/function; ≤ 20 LOC; CC ≤ 5 |
| **DRY** | Duplication? Extract |
| **Simplicity** | Choose the simpler path |
| **Explicit > Implicit** | Self‑documenting, or add comments |
| **Fail fast** | Clear, early error handling |
## 2 Architecture
### Files / Packages
- ≤ 400 logical LOC
- No “util” dumping grounds
- Naming:
- `ComponentName.tsx` (PascalCase)
- `moduleName.ts` / `moduleName.css` (camelCase)
### Modules & Dependencies
1. Each module owns **one** domain concept.
2. Export only what callers need (`index.ts`).
3. No import cycles – break with interfaces.
4. Import depth ≤ 3.
5. Prefer composition; inherit only if ≥ 2 real subclasses.
6. Keep domain pure; use **Ports & Adapters** for I/O.
7. Names: packages/modules = nouns; functions = verb + noun.
## 3 Testing
- **One** happy‑path unit test per public function (unless CC > 5).
- Integration tests only at service seams; mock internals.
---
## 🤔 When You're Unsure
1. **Stop** and ask as many questions as necessary to remove any ambiguity or clarify any knowledge gaps.
2. Use the AskUserQuestion tool
## Troubleshooting
If you encounter issues:
- Use perplexity for research (the current year is 2026)
- If you need clarification, ask specific questions with AskUserQuestion tool
