Get Started
Getting Started with Galdr
Install galdr into any existing project in under 5 minutes. No external services required — galdr is entirely file-based and works with any IDE that supports AI agents.
Prerequisites
- ✓ Git installed and configured
- ✓ An existing project (any language, any framework)
- ✓ At least one supported IDE: Cursor, Claude Code, Gemini CLI, OpenAI Codex, or OpenCode
Installation
Clone the galdr repository, then copy the template files into your project root.
Clone the galdr repository
Clone galdr somewhere on your machine — you only need to do this once per machine.
git clone https://github.com/wrm3/galdr.git ~/galdrCopy the template into your project
Copy the galdr system files into your project root. This adds the .galdr/, .claude/, and .cursor/ directories.
# macOS / Linux
cp -r ~/galdr/template_full/. /path/to/your-project/
# Windows (PowerShell)
Copy-Item -Recurse -Force "$HOME\galdr\template_full\*" "C:\path\to\your-project\"template_full for everything including the vault and PCAC system. If you want the minimal version, use template_slim instead.Run galdr setup
Open your project in your AI IDE and run the setup command. This initializes your .galdr/PROJECT.md, generates initial goals, and sets your user identity.
# In your AI IDE chat:
@g-setupThe agent will ask you a few questions about your project (name, mission, tech stack) and populate the initial galdr files.
Verify the installation
Run the status command to confirm galdr loaded correctly. You should see your project name, goals, and an empty task queue.
@g-statusExpected output: your project name, goals, active phase, idea count, constraint count, and "No active tasks."
Your First Task
Now let's create a real task, implement it, and verify it — the complete galdr loop.
Create a task
Use @g-task-new to create a properly-specced task with acceptance criteria.
@g-task-new Add a health-check endpoint to the APIGaldr will create a task file at .galdr/tasks/task001_*.md with an ID, priority, subsystem, and acceptance criteria. Review and edit the file if needed.
Implement the task
Run the implementation command. The agent reads all open tasks, implements them in priority order, checks each acceptance criterion, and marks completed items [🔍] (awaiting verification — NOT done yet).
@g-go-code[🔍]. This is galdr's adversarial quality gate.Verify the task
Open a new agent session and run the review command. A structurally separate agent checks the implementation against the acceptance criteria. Only this agent can mark work [✅].
# In a NEW chat session:
@g-go-reviewIf everything passes: your task is [✅] complete. If something fails: it goes back to pending with a FAIL note — and you repeat from step 6.
Commit the work
@g-git-commitGenerates a structured commit message with the correct type prefix (feat:, fix:, etc.), task reference, and optional agent footer.