https://github.com/g70245/cg

A Windows desktop automation tool built with Go, Fyne, and the Win32 API. It provides visual workflow editing, conditional state-machine execution, multi-instance coordination, and game-state detection through process memory, pixel sampling, and log monitoring.

Developed as a personal side project since 2023. A vendor-neutral AI-assisted development workflow was introduced in 2026.  

Key Features and Engineering Highlights

1. Visual Combat Workflow Editor

Designed a graphical editor for configuring ordered character and pet actions without writing code. Each action can define conditions such as enemy position, enemy count, health percentage, skill parameters, and target selection. 

2. Conditional State Machine

Implemented a configurable finite-state machine that evaluates action outcomes and determines whether to:

- Continue to the next action - Jump to a specified action

- Retry the current flow

- Restart from the beginning

- Switch between character and pet actions

This enables flexible combat workflows instead of fixed action sequences.

3. Windows and Game-State

Integration Combined multiple data sources to determine game state:

- Win32 keyboard and mouse messages

- Window discovery and multi-instance handling

- Pixel and color sampling at fixed client coordinates

- Process-memory reads for map names and character coordinates

- Big5-encoded game-log parsing

Added explicit process-handle cleanup and error-aware file handling to improve long-running stability. 

4. Movement Strategies

Used process-memory coordinates, map information, trigonometric calculations, and coordinate transformations to implement configurable movement strategies. Movement modes support direction, reversal, positional bias, and timing adjustments while remaining constrained by the active workflow. 

5. Multi-Instance Coordination

Used goroutines, channels, tickers, and shared workflow state to coordinate multiple game windows. Supported capabilities include:

- Grouped combat workflows

- Per-window action configuration

- Teleport and resource monitoring

- Health, mana, inventory, and verification checks

- Semi-automated production and inventory organization

6. Operator Alerts

Implemented repeating MP3 alerts for conditions requiring manual intervention, including:

- Low health or mana

- Full inventory

- Resource depletion

- Teleport detection

- Verification messages

- Production failures

7. Build, Testing, and CI

Created a repeatable Windows development workflow with:

- PowerShell build and packaging scripts

- Fyne release packaging

- GitHub Actions CI

- Automated Go tests and static analysis

- Table-driven tests for business logic, process-handle ownership, filesystem failures, and log validation 

8. Vendor-Neutral AI-Assisted Development Workflow

Established a vendor-neutral AI-assisted development workflow in 2026.

The repository serves as the single source of truth for specifications, task status, engineering decisions, workflow rules, and validation evidence. Different AI tools use lightweight adapters while following the same processes for context loading, planning, execution authorization, branch safety, validation, commit review, and cross-session handoff.

This approach reduces dependency on a specific AI platform while keeping AI-assisted changes portable, traceable, reviewable, and subject to human approval.  

Technologies

Go, Fyne, Win32 API, CGO, Goroutines, Channels, Finite-State Machines, Process Memory, Pixel Detection, Big5 Log Parsing, PowerShell, GitHub Actions, Table-Driven Testing, Vendor-Neutral AI-Assisted Development, Human-in-the-Loop Engineering, OpenAI Codex 

Project Scope

Personal technical research project developed since 2023 and used only in an unofficial third-party server testing environment. It has not been used commercially.