Repository Organization
Relevant source files
The following files were used as context for generating this wiki page:
The repository is structured as a collection of specialized system prompts and model configurations for various AI coding tools and agents. Its primary purpose is to serve as a centralized reference for the operational instructions, tool definitions, and architectural constraints that govern different agentic IDEs and AI assistants. The organization follows a vendor-specific or tool-specific directory hierarchy, facilitating the management of diverse prompt engineering strategies ranging from simple chat instructions to complex agentic workflow specifications.
Sources: README.md, Amp/gpt-5.yaml:385-425
Directory Hierarchy and Categories
The project organizes assets based on the specific AI tool or service provider. Each top-level directory typically contains text-based instructions (system prompts) or configuration files (YAML) that define the behavior, identity, and toolset of a specific AI agent.
Core Tool Categories
The repository contains specialized prompts for both proprietary and open-source agents:
- Proprietary Agents: High-level directories for specialized tools like
Google/Antigravity,Amp,Cursor Prompts, andDevin AI. - Open Source Frameworks: Dedicated sections for community-driven or open-source tools such as
Bolt,Codex CLI, andGemini CLI. - Platform Specifics: Configurations for cloud-based IDEs and deployment platforms like
Same.dev.
Sources: Amp/gpt-5.yaml:390-420, README.md
Structural Overview
The following diagram illustrates the high-level organizational structure of the repository's directories.
Component Architecture and Artifact Management
The repository defines how agents should interact with specific file structures within the user's workspace. Many agents are instructed to maintain hidden or specific directories to track state, tasks, and documentation.
Agentic State Directories
Several prompts define a "brain" or state directory used to store living documents such as task lists and implementation plans. For instance, the Antigravity agent uses a specific path for its internal state:
| Component | Path Pattern / Requirement | Description |
|---|---|---|
| Brain Directory | ~/.gemini/antigravity/brain/{UUID}/ | Stores task.md and implementation_plan.md for state tracking. |
| Quest Directory | .qoder/quests/ | Stores design documents generated during the "Quest" workflow. |
| Memo Directory | .same/ | Contains todos.md, wikis, and documentation for internal agent tracking. |
| Workflow Dir | .agent/workflows/ | Stores markdown-based step-by-step instructions for specific tasks. |
Sources: Google/Antigravity/planning-mode.txt:46-75, Qoder/Quest Design.txt:105-110, Same.dev/Prompt.txt:64-75
Data Flow for Task Execution
The repository documentation details how information flows from a user query through planning artifacts to final implementation.
Configuration and Toolset Specification
Individual files within the repository act as a manifest for tool capabilities. Each agent configuration specifies available functions, such as file system manipulation, shell execution, and web searching.
YAML Configuration Pattern (Amp)
The Amp/gpt-5.yaml file demonstrates a structured approach to defining the environment, including operating system details, working directories, and a cached directory listing to minimize redundant ls calls.
Sources: Amp/gpt-5.yaml:385-425
Tool Schema and Limitations
Prompts often include strict constraints on how the file system can be accessed. For example, the Bolt prompt specifies that it operates within a WebContainer environment, which lacks native binary support and git capabilities.
Sources: Open Source prompts/Bolt/Prompt.txt:3-20
Summary of Repository Significance
The organization of this repository centralizes the "intelligence" of various AI coding assistants by defining their operational boundaries. By separating prompts into specialized modules (Planning, Execution, Quest Design), it allows for a granular approach to software engineering tasks where different models or instructions are invoked based on the current stage of the development lifecycle. This structure ensures that agents maintain technical accuracy and follow consistent patterns for task management and verification.
Sources: Google/Antigravity/planning-mode.txt:14-25, Qoder/Quest Design.txt:80-85