How heckenshutze/https-github.com-x1xhlol-system-prompts-and-models-of-ai-tools works
This repository is a static corpus of AI assistant prompts and tool definitions, not an executable application. Its architecture is organized around vendor-specific prompt bundles and a small amount of repository-level metadata.
1. Repository role
- The root contains a README.md, a LICENSE.md, and a top-level .gitignore.
- There is no application runtime, service layer, or build pipeline visible in the tree; the “product” is the curated collection itself.
2. Primary organizing principle: one folder per AI tool/vendor
- Most top-level directories map to a specific AI product or agent environment: Cursor Prompts, Windsurf, Claude Code, Replit, Trae, Xcode, Qoder, Lovable, Augment Code, etc.
- This creates a clear boundary between systems: each folder is an isolated prompt/tool package for one assistant family.
3. Artifact types inside each vendor folder
- The contents are mostly plain-text prompt files (.txt) and, where relevant, tool manifests (.json).
- These files represent the assistant’s operating instructions, chat prompts, agent prompts, and sometimes memory or mode-selection prompts.
4. Prompts vs. tools are the two core data classes
- Prompt files encode behavior and policy: system prompts, chat prompts, agent prompts, task-specific prompts, memory prompts, and decision prompts.
- JSON files encode callable tools or tool schemas for agents that can interact with external actions, files, or IDE features.
5. Examples of vendor-specific packaging
- Cursor Prompts contains multiple versions of agent prompts plus tool definitions and memory-related prompts.
- Claude Code includes a system prompt and tools JSON, reflecting a code-focused agent runtime.
- Replit, Same.dev, Trae, Lovable, Augment Code, and Windsurf each pair prompts with tool metadata where applicable.
6. Specialized substructures
- Some vendors expose multiple operating modes: for example Kiro has classifier/spec/vibe prompts, suggesting a mode-routing layer before task execution.
- Xcode is split by action type (ExplainAction, MessageAction, PreviewAction, etc.), indicating UI/editor-command integration rather than a single generic prompt.
7. Open-source assistant bucket
- Open Source prompts/ is a distinct subtree that groups tools with public implementations such as Codex CLI, Cline, Bolt, RooCode, Lumo, and Gemini CLI.
- This section acts like a comparative archive of open ecosystems, separate from vendor-specific products.
8. Repository-level navigation and discovery
- The README.md is the catalog: it lists available assistants, gives update notes, and links out to community/support channels.
- The repo’s “front door” is documentation-first; consumers are expected to browse by vendor folder rather than discover through code references.
9. Versioning is file-based, not code-based
- There is no central schema registry or package manager; evolution is represented by multiple prompt variants in filenames, such as “v1.0”, “v1.2”, “Wave 11”, or dated agent prompts.
- That means architectural changes are captured as new prompt artifacts alongside older ones, preserving historical snapshots.
10. External dependencies and infrastructure
- The repository references external platforms only in documentation and badges: Discord, Trendshift, DeepWiki, Cloudback, and payment/support services.
- These do not affect the runtime architecture of the repo, but they do define the project’s distribution and community layer.
11. What to diagram
- A useful architecture diagram would show:
- the repository root as a catalog/index,
- vendor folders as isolated prompt bundles,
- prompt text files as the behavioral layer,
- JSON tool files as the action/tooling layer,
- and the open-source subcollection as a separate comparative branch.
12. Overall architectural character
- This is best modeled as a content library / reference archive for AI agent configurations.
- The main boundary is not between services or modules, but between different assistant ecosystems and their prompt/tool contracts.
Main entry points: Cursor Prompts Cursor Prompts · README.md README · Claude Code Claude Code · Windsurf Windsurf
Components 18
Repository root
Top-level catalog, metadata, and documentation.
Vendor bundles
Isolated prompt/tool packages for specific assistant ecosystems.
- Cursor Prompts · Cursor-specific prompts, memory, and tool definitions.
Cursor Prompts - Claude Code · Code-focused system prompt and tools contract.
Claude Code - Windsurf · Wave-based prompt and tool package.
Windsurf - Replit · Prompt plus tool schema for Replit.
Replit - Trae · Builder and chat prompts with tool metadata.
Trae - Xcode · Action-specific prompts for IDE and preview workflows.
Xcode - Kiro · Classifier, spec, and vibe prompts for mode routing.
Kiro - Lovable · Agent prompt and tool definitions.
Lovable - Augment Code · Model-specific prompts paired with tools JSON.
Augment Code - Manus Agent · Loop, modules, prompt, and tools package.
Manus Agent Tools & Prompt - Other vendors · Additional single-folder assistant prompts such as Devin, Junie, Cluely, Orchids, Perplexity, VSCode Agent, Warp.dev, dia, Qoder, Same.dev, and v0.
Open-source archive
Comparative collection of public assistant prompt implementations.
- Open-source prompts · Public assistant prompt collection grouped by open ecosystems.
Open Source prompts - Open implementations · Bundled open-source assistants such as Codex CLI, Cline, Bolt, RooCode, Lumo, and Gemini CLI.
Other components
- Prompt files · Text prompts that define agent behavior, policy, and modes.
- Tool schemas · JSON tool manifests describing callable actions and integrations.
- Versioned variants · Multiple prompt revisions preserved as sibling files instead of centralized code.
Connections 26
- README → Cursor Prompts: indexes · README catalogs the assistant bundles.
- README → Open-source prompts: indexes · README also points to the open-source collection.
- Cursor Prompts → Prompt files: contains · Cursor ships prompt variants and memory prompts.
- Cursor Prompts → Tool schemas: contains · Cursor includes tool definitions as JSON.
- Cursor Prompts → Versioned variants: evolves via · Multiple prompt versions coexist in the folder.
- Claude Code → Prompt files: contains · Claude Code centers on a system prompt.
- Claude Code → Tool schemas: contains · Claude Code exposes a tools contract.
- Windsurf → Prompt files: contains · Windsurf provides prompt waves.
- Windsurf → Tool schemas: contains · Windsurf pairs prompts with tool instructions.
- Replit → Prompt files: contains · Replit package includes a prompt file.
- Replit → Tool schemas: contains · Replit package includes tool schemas.
- Trae → Prompt files: contains · Trae includes builder and chat prompts.
- Trae → Tool schemas: contains · Trae includes builder tools.
- Xcode → Prompt files: contains · Xcode action files are specialized prompts.
- Kiro → Versioned variants: routes by mode · Classifier/spec/vibe prompts define distinct modes.
- Lovable → Prompt files: contains · Lovable includes agent and generic prompts.
- Lovable → Tool schemas: contains · Lovable includes a tool manifest.
- Augment Code → Prompt files: contains · Augment bundles model-specific agent prompts.
- Augment Code → Tool schemas: contains · Augment bundles matching tool JSON.
- Manus Agent → Prompt files: contains · Manus uses loop, module, and prompt files.
- Manus Agent → Tool schemas: contains · Manus includes tool schemas.
- Other vendors → Prompt files: contains · Other vendor folders are mostly standalone prompt bundles.
- Open-source prompts → Open implementations: groups · Open-source subtree collects public implementations.
- Open implementations → Prompt files: contains · Each open implementation is represented by prompt text.
- Prompt files → Versioned variants: includes · Prompt behavior is versioned through filename variants.
- Tool schemas → Versioned variants: includes · Tool definitions may also appear in versioned forms.