A spec-driven development toolkit to eliminate AI code bloat and automate software lifecycle tasks.
💡 Foundational Philosophy: Read about our philosophy in the blog post: Beyond the Prompt: A Practical Guide to Context Engineering with Gemini.
Context Engineering 2.0 represents a complete paradigm shift from raw, conversational prompts to structured software engineering workflows. Instead of "vibe-coding" and hoping the LLM gets it right, Context Engineering structures the context, templates, rules, and commands natively within the Antigravity CLI (agy) to ensure consistent, reliable, and production-grade code generation.
| Feature | Vibe Coding (Prompt Engineering 1.0) | Context Engineering 2.0 (Antigravity Edition) |
|---|---|---|
| Workflow | Sending raw, unstructured prompts in chat. | Executing compiled, structured specifications (PRPs). |
| Code Quality | Prone to over-engineered blocks and dependency bloat. | Minimalist, spec-driven code restricted by toggleable guidelines (Ponytail mode). |
| Interaction | Constant conversational fluff and back-and-forth. | Native, non-interactive action blocks and precise file diffs. |
| Context Management | Context drift as the conversation grows. | Unified, project-specific rulesets and slash commands natively registered in agy. |
Install the unified plugin natively in your terminal with a single command:
agy plugin install https://github.com/Apoo711/Context-EngineeringThe suite registers two native slash commands directly into your agy environment to manage the lifecycle of your features:
Generates a rigid, comprehensive Product Requirements Prompt (PRP) template based on a simple user request. It forces the AI to plan edge cases, testing gates, and structures before writing a single line of code.
Executes a compiled PRP spec sheet. It uses strict, non-interactive action blocks to cleanly modify or create files without conversational fluff.
Tired of AI agents pulling down heavy third-party libraries for trivial tasks? Ponytail Mode is our built-in minimalist coding guidelines ruleset.
The philosophy is simple: "Lazy, not negligent." Ponytail mode forces the agent to exhaustively evaluate standard libraries and native browser features before introducing external dependencies.
You can configure Ponytail Mode by editing the enable_ponytail_mode setting in your plugin.json:
{
"settings": {
"enable_ponytail_mode": true
}
}Alternatively, you can toggle it interactively through the visual /config menu in agy or update it in your global settings.
Here is where the Context Engineering Suite is heading next:
- Progressive Agent Personas: Toggle between specialized agent profiles (e.g., Greenfield Builder vs. Hardened-Security Auditor).
- Auto-Verification Test Hooks: Post-invocation validation gates that automatically run your test suites to verify generated code correctness.
- Visual Workspace Indexing (
/context-map): Auto-generate interactive dependency and architecture maps directly in your terminal.
This project's workflow and foundational concepts were inspired by the original Context-Engineering-Intro repository by coleam00. While the codebase has transitioned into a native Antigravity CLI plugin, the core vision of spec-driven AI development remains.