---
name: ecolab-ecodata-research
description: Autonomous quantitative economics research skill for Antigravity IDE connecting EcoLab (https://ecolab.io.vn) and EcoData MCP (https://ecodata.io.vn) within the TNSAI ecosystem (https://www.tnsai.vn).
---

# Antigravity IDE Skill — EcoLab & EcoData MCP Quantitative Economics Suite

## 1. Ecosystem Architecture
- **TNSAI (`https://www.tnsai.vn`)**: Parent scientific research and economic intelligence ecosystem.
- **EcoLab (`https://ecolab.io.vn`)**: 5-stage quantitative research workbench (`Ideation`, `Literature Review`, `Data Collection`, `Modeling`, `Reporting`).
- **EcoData (`https://ecodata.io.vn`)**: Empirical economic data catalog (14,000+ Vietnam & international series) and MCP server.

## 2. Antigravity Agent Protocol (Host-LLM-First & Safe Tool Execution)
0. **Host-LLM-First Mode (`X-EcoLab-Execution-Mode: host_llm`)**:
   - Prioritize using the **Host LLM (e.g., Gemini 4 Argon + Deep Thinking)** to reason, synthesize literature, interpret econometric diagnostics, and write complete academic reports directly from the structured empirical data, regression tables, and Stata/R/Python scripts returned by EcoData MCP and EcoLab tools.
   - Never invoke server-side external LLM Provider API keys during MCP or Agent tool execution (external LLM Provider API keys are only used when interacting directly on the EcoLab Web App `https://ecolab.io.vn`).
   - Comply strictly with MCP Prompt Injection & Safe Tool Execution guardrails (clean variable identifiers `^[A-Za-z_][A-Za-z0-9_.]*$`, zero shell/system/SQL/prompt injection in tool arguments).
1. **Zero Synthetic Data / Zero Fake Citations**:
   - Never fabricate author names, publication years, DOIs, descriptive statistics, or regression coefficients.
   - All literature citations must originate from verified queries across Semantic Scholar, OpenAlex, CrossRef, CORE, DOAJ, or ArXiv.
2. **EcoData MCP Data Retrieval**:
   - Query `vietnam_macro_gso_nso_query`, `vietnam_macro_customs_trade_query`, `vietnam_macro_fiscal_mof_query`, `vietnam_macro_survey_index_query`, and `global_macro_world_query`.
   - Align frequency, currency/unit definitions, and panel identifiers (`province_code`/`country_iso3`, `year`) before modeling.
3. **Diagnostic-First Econometric Modeling**:
   - Always run and report multicollinearity (`VIF`), heteroskedasticity (`Breusch-Pagan` / `White` / `Modified Wald`), autocorrelation (`Breusch-Godfrey` / `Wooldridge`), stationarity (`ADF` / `PP` / `KPSS` / `IPS` / `CIPS`), and endogeneity (`Durbin-Wu-Hausman` / `Hansen J`).
4. **Triple-Language Reproducibility**:
   - Every empirical analysis must produce synchronized **Stata (`.do`)**, **R (`.R`)**, and **Python (`.py`)** scripts so researchers can replicate results independently.