Meridian provides a suite of interactive **Agent Skills** designed to assist
data scientists and engineers throughout the Marketing Mix Modeling (MMM)
lifecycle.

These skills follow the open
[Agent Skills specification](https://agentskills.io) and provide AI coding
assistants with domain-specific expertise to help you prepare data, configure
priors, fit models, analyze diagnostics, run budget optimizations, and plan
scenarios.

*** ** * ** ***

## What are agent skills?

Agent Skills are modular, standardized capability packages containing
instructions, reference templates, and scripts. When integrated into an AI
coding assistant, skills enable the agent to:

- **Guide you step-by-step**: Navigate complex statistical configuration without needing to memorize the entire API.
- **Infer schema mappings**: Analyze raw CSV column headers and suggest appropriate mappings for KPIs, media spend, impressions, reach/frequency, and control variables.
- **Maintain human-in-the-loop control**: Present configuration proposals (such as Bayesian priors, knot selection, and spend constraints) at critical checkpoints and await your confirmation before execution.
- **Generate reproducible Python scripts** : Produce standalone Python scripts runnable directly in standard Python virtual environments (`python3 script.py`).

*** ** * ** ***

## Available Meridian skills

Meridian includes five specialized skills covering each stage of the MMM
journey:

| Skill | Identifier | Description | Key Outputs |
|---|---|---|---|
| **Documentation Consultant** | `meridian-doc-consultant` | Answers conceptual, mathematical, and configuration questions by citing relevant Meridian documentation. | Grounded technical explanations \& references |
| **Model Building** | `meridian-model-building` | Guides you through loading CSV data, column mapping, `ModelSpec` \& prior setup, exploratory data analysis (EDA), MCMC fitting, and model saving. | Python script, EDA HTML report, saved model file (`.binpb`) |
| **Result Visualization** | `meridian-result-visualization` | Loads fitted models to generate MCMC convergence health checks (`Reviewer`) and 2-page executive summary reports (`Summarizer`). | Health check HTML, Summary results HTML |
| **Budget Optimization** | `meridian-budget-optimization` | Configures and runs fixed/flexible budget optimizations and target ROI spending scenarios. | Budget optimization report HTML |
| **Scenario Planner** | `meridian-scenario-planner` | Prepares scenario planner data and provides handoff instructions to create Looker Studio dashboards. | Scenario planner data file (`.binpb`), Dashboard URL |

*** ** * ** ***

## Installation and discovery

Meridian skills are distributed directly in the
[open-source repository](https://github.com/google/meridian) and can be added to
your workspace using standard agent tooling.

### Install using the Skills CLI

If you have Node.js / `npm` installed in your environment, you can install all
Meridian skills or specific individual skills using `npx skills`:

    # Install all Meridian skills into your workspace
    npx skills add google/meridian
    # Or install a specific skill (e.g., model building)
    npx skills add google/meridian --skill meridian-model-building

### Manual or in-repository discovery

If you prefer not to use `npx` or are working in a pure Python environment:

- **In-repository discovery** : If you clone or develop directly within the `google/meridian` repository, compatible AI coding assistants automatically discover all available skills using the workspace manifest at `.agents/skills.json` and the `skills/` directory.
- **Manual copy** : You can copy the `skills/` subfolders (e.g., `meridian_model_building/`) from the repository directly into your project's `.agents/skills/` directory.

*** ** * ** ***

## Supported environments

Meridian skills are built on the open
[Agent Skills specification](https://agentskills.io). Because they use open
standards (standard Markdown instructions and modular Python templates), they
are designed to be compatible with any agentic coding environment, IDE
extension, or CLI tool that supports the Agent Skills standard.

> [!NOTE]
> **Note:** Meridian skills are designed to generate standard Python 3 code that runs in your local Python virtual environment (`python3 your_script.py`) using standard PyPI packages (`google-meridian`). No proprietary internal tools or monorepo environments are required.

*** ** * ** ***

## Human-in-the-loop safety and verification

Meridian skills follow a **checkpoint-driven design**. During interactive
sessions, the agent will pause and prompt for your review before:

1. Finalizing column mappings for your dataset.
2. Setting Bayesian prior parameters (such as ROI priors or knot spacing).
3. Running long MCMC posterior sampling jobs.
4. Executing generated scripts against your local environment.

This ensures you retain full visibility and control over your modeling
decisions while benefiting from automated script authoring.