- Home
- Skills
- Prompt & Skill Engineering
- AntiGravity Starter Kit — Context Optimizer & Lean Core
More screenshots
Works with the AI Agents you already use
AntiGravity Starter Kit — Context Optimizer & Lean Core
Eliminates context bloat, audits rule files for token bleed, and enforces the under-40-line Lean Core standard.
$26
AntiGravity Starter Kit — Context Optimizer & Lean Core
Example session with this skill installed
Audit my 450-line AGENTS.md and .cursorrules file for context bloat, identify token waste per conversation turn, and convert it to the Lean Core standard.
- Read your context and instructions
- Compiled the antigravity starter kit
Rule File Token Audit
- Original Root Rule Size: 462 lines (14,800 tokens per turn)
- Attention Degradation Risk: 🔴 HIGH (Lost-in-the-Middle trigger detected)
- Token Bleed Projection: ~296,000 wasted tokens over a 20-step debugging session
Refactored Lean Core Architecture:
- Root
AGENTS.md: Stripped down to 24 lines of universal invariants (720 tokens). - Modular References (
references/): 4 specialized markdown files created for on-demand loading. - Subagent Router (
rules/model_routing.md): Configured for automatic task delegation.
# AGENTS.md — Lean Core Standard (<30 lines)
1. Invariant: Maintain 100% test pass rate before committing code.
2. Routing: Load specialized references on-demand using view_file.
3. Subagents: Heavy refactoring runs in isolated subagents (context: fork).
Connects securely to your tools. The creator never sees your data.
About this skill
AntiGravity Starter Kit — Context Optimizer & Lean Core Rules
Tired of your AI coding assistant forgetting earlier instructions, burning tokens at lightning speed, or drifting into hallucinations?
The AntiGravity Starter Kit diagnoses and solves Context Bloat across Claude Code, Cursor, Windsurf, OpenClaw, Codex CLI, and Google Antigravity.
⚠️ The Problem: Monolithic Instructions Degrade LLMs
When your root rule files (AGENTS.md, .cursorrules, CLAUDE.md) exceed 80–100 lines:
- Attention Degradation (Lost-in-the-Middle): The LLM dilutes constraints, causing the assistant to violate safety rules or repeat mistakes.
- First-Token Latency (TTFT) & Token Waste: Every turn re-reads the full system prompt. Over 125,000 wasted tokens in a 50-step session.
- Premature Context Overflow: Heavy instructions push the session into early context exhaustion.
⚡ The Solution: The AntiGravity 3-Tier Lean Core (<40 lines)
- Tier 1: Lean Root Core (<40 lines): Universal behavioral invariants only.
- Tier 2: Modular Reference Modules: Domain knowledge loaded on-demand from
references/. - Tier 3: Isolated Subagents (
context: fork): Heavy refactoring in sandboxed sub-windows.
📦 What's Included:
scripts/token_inspector.py(CLI diagnostic audit)assets/AGENTS.template.md(18-line production starter)references/(Architecture guides and recipes)agents/openai.yaml(Codex CLI metadata)
🔒 100% Local-First & Zero Telemetry
Runs entirely on your local machine using your existing CLI environment. Zero data telemetry, zero cloud egress.
Focuses on instruction optimization. Does not audit application code performance, database queries, or cloud infrastructure costs.
How to use
- 1
Connect your agent to the Agensi MCP
Copy the prompt and paste it into Claude, ChatGPT, Cursor or any agent that supports MCP. It sets up the connection for you. Manual setup
- 2
Sign in to Agensi
Your agent shows a sign-in link. Sign in and click Allow.
- 3
Ask for what you need
Your agent loads this skill and uses it on your task. It always asks before any payment.
Reviews
No reviews yet
Be one of the first to try it. Every listed skill goes through the trust checks below.
Security scanned
Checked by our 8-point scan before listing
Fresh listing
Recently published to Agensi
Money-back guarantee
Doesn't do what its page promises? You get your money back.
Trust & safety
Security scan
Security scan passed 1 month ago
Requires
Permissions
Required to audit local rule files and refactor rule structures into modular directories.