Legacy Ai Chart Agent

---
name: legacy-ai-chart-agent
description: "Maintain and inspect the deprecated vizro-ai chart agent, response
  models, and tests without using it for new greenfield workflows."
disable-model-invocation: true
metadata:
  disco-role: operating
license: Apache 2.0
---

# Legacy AI Chart Agent

Use this sub-skill when the task explicitly mentions `vizro-ai`, the chart agent, chart-agent tests, response models, safeguards, or legacy AI chart-generation behavior.

For new agent-assisted dashboard work, route to `../mcp-agent-workflows/SKILL.md` instead unless the user specifically asks for `vizro-ai`.

## Deprecation facts

- `vizro-ai` is deprecated in this checkout. Importing `vizro_ai` emits a package deprecation warning that points users toward Vizro e2e-flow or Vizro-MCP.
- There is no top-level `VizroAI` class exported from `vizro_ai.__init__` in this snapshot.
- The active chart agent evidence is in `vizro_ai.agents._chart_agent.chart_agent`.

## Current chart-agent surface

```python
import pandas as pd
from vizro_ai.agents._chart_agent import add_df, chart_agent
from vizro_ai.agents.response_models import BaseChartPlan

# chart_agent is a pydantic-ai Agent that expects a pandas DataFrame dependency.
```

`add_df(ctx)`:

- Requires `ctx.deps` to be a pandas `DataFrame`.
- Samples five rows and appends `DataFrame.info()` text to the agent instructions.
- Raises `ValueError` when dependency is not a DataFrame.

## Safe maintenance workflow

1. Keep deprecation messaging intact unless the task is explicitly to change deprecation policy.
2. Avoid live model/provider calls for routine tests. Unit-test helpers, response models, and safeguards with local DataFrames.
3. If provider-backed tests/examples are requested, confirm API keys, model choices, network permission, and cost/budget first.
4. For chart code generated by the agent, run/validate the generated Plotly code before embedding it in a Vizro dashboard.

## Focused tests

From `vizro-ai/`:

```bash
hatch run test-unit tests/unit/vizro-ai/agents/test_chart_agent.py
hatch run test-unit tests/unit/vizro-ai/test_init.py
```

If using a prepared environment instead of Hatch, run equivalent `pytest` from `vizro-ai/` after local package install.

## Common pitfalls

- Treating `vizro_ai` as a public high-level SDK: it is deprecated and narrow in this snapshot.
- Importing a non-existent top-level `VizroAI` class.
- Passing non-DataFrame deps into chart-agent instruction helpers.
- Allowing live provider calls in tests without explicit user authorization.
- Forgetting that new dashboard generation should usually use Vizro-MCP or e2e-flow guidance instead.

## Evidence anchors

- `vizro-ai/src/vizro_ai/__init__.py`: deprecation warning and public package initialization.
- `vizro-ai/src/vizro_ai/agents/_chart_agent.py`: `chart_agent`, `add_df`.
- `vizro-ai/src/vizro_ai/agents/response_models.py`: chart response models.
- `vizro-ai/tests/unit/vizro-ai/agents/test_chart_agent.py`: behavior expectations.
- `vizro-ai/docs/pages/API-reference/vizro-ai.md` and `vizro-ai/examples/example.py`: public/deprecated documentation context.