System Overview

PREQSTATION is a task management platform designed for AI-agent-driven development workflows. It bridges developers and AI agents through a central task queue and REST API.

Three-System Architecture

PREQSTATION consists of three interconnected services:

projects-manager

Central web app. Manages tasks, renders the Kanban UI, exposes the REST API, and can relay Telegram/OpenClaw messages.

preqstation-openclaw

Execution skill package. OpenClaw imports it to resolve projects, create worktrees, and launch coding agents.

preqstation-skill

Agent client. MCP server and shell helpers for agents to call projects-manager APIs during execution.

System Diagram

┌──────────────────────────────────────────────────────────────┐
│                  projects-manager (Web App)                  │
│                                                              │
│  Next.js 16 · React 19 · Mantine 8 · Drizzle · Neon PG     │
│                                                              │
│  ┌────────────┐  ┌──────────────┐  ┌───────────────────┐    │
│  │ Kanban UI  │  │  REST API    │  │  Telegram Client  │    │
│  │ Dashboard  │  │  /api/tasks  │  │ /api/telegram/*   │    │
│  │ Settings   │  │ /api/api-keys│  │                   │    │
│  └────────────┘  └──────┬───────┘  └────────┬──────────┘    │
│                         │                    │               │
└─────────────────────────┼────────────────────┼───────────────┘
                          │                    │
                  API calls (Bearer token)     │ Sends notifications
                          │                    │
                          │                    ▼
                          │           ┌─────────────────┐
                          │           │    Telegram     │
                          │           └────────┬────────┘
                          │                    │
                          │                    │ User message / bot relay
                          │                    ▼
                          │           ┌─────────────────────────────────┐
                          │           │      OpenClaw (Runtime)         │
                          │           │                                 │
                          │           │  Receives request, imports      │
                          │           │  preqstation-openclaw skill     │
                          │           └────────────────┬────────────────┘
                          │                            │
                          │                   Parses intent, creates
                          │                   git worktree, launches
                          │                   coding agent (PTY+bg)
                          │                            │
                          │                            ▼
                          │           ┌─────────────────────────────────┐
                          │           │     Coding Agent                │
                          │           │ ("claude-code" / "codex" /      │
                          │           │   "gemini-cli")                 │
                          │           │                                 │
                          │           │  Runs in isolated worktree      │
                          │           │  Implements code changes        │
                          ◄───────────│  Uses preqstation-skill MCP     │
                                      │  to report status & results     │
                                      └─────────────────────────────────┘

Data Flow

Task Creation and Notification

User creates task in Kanban UI
  → projects-manager stores in PostgreSQL
  → (optional) sends Telegram notification

Agent Trigger via OpenClaw

Developer sends task request to OpenClaw
  → OpenClaw runtime receives message
    → preqstation-openclaw skill triggers
      → Parses: engine key, task ID, project key
      → Resolves project path from agent memory
      → Creates isolated git worktree
      → Launches coding agent (claude/codex/gemini CLI)

Agent Execution and Status Reporting

Coding agent fetches task via preq_get_task, then:

  inbox       → preq_start_task (run_state → working)
              → preq_plan_task (write plan, → todo, clear run_state)
  todo/hold   → preq_start_task (run_state → working)
              → Implement + test + deploy per strategy
              → preq_complete_task (→ ready, clear run_state)
  ready       → preq_start_task (run_state → working)
              → Run verification (tests, build, lint)
              → preq_review_task (→ done, clear run_state)

  Telegram dispatch before pickup may set run_state=queued
  On any failure → preq_block_task (→ hold)

Tech Stack

Layer	Technology
Framework	Next.js 16 (App Router)
UI	React 19, Mantine 8, Tabler Icons, Recharts
ORM	Drizzle ORM
Database	Neon PostgreSQL (Row Level Security)
Validation	Zod
Auth	Session cookies (HMAC-signed), bcryptjs