Agent Skill — AVA Docs

Give your agent data analysis tools, from the first question to a shareable chart.

START WITH YOUR AGENT

Install with your agent

One prompt to get ready. Your agent checks the environment, installs AVA and the skill, and verifies that everything works.

Paste into your terminal-based AI agent.

View installation prompt

Install AVA's Agent Skill in the current project so you can analyze data and create charts for me.
 
1. Check that this environment runs macOS or Linux, with Node.js 22.13 or later and npm available. Use an existing Node version manager if an upgrade is needed. If you cannot meet these requirements, explain what is missing.
2. Install the CLI with `npm install -g @antv/ava@alpha`. Verify that `ava --help` and `ava source --help` work.
3. Install the skill for the detected agents in this project with `npx --yes skills add antvis/AVA --skill ava --yes`.
4. Read the installed ava/SKILL.md and its environment setup reference. Verify the installed skill with `npx --yes skills list`.
5. Smoke-test the CLI using a tiny CSV in a new temporary directory: load it with `ava source`, inspect it with `ava schema`, and execute a simple aggregate with `ava query`. Use the returned dataset ID and the actual table name from the schema. Dispose of this test session afterward.
 
Use your own reasoning for analysis; do not ask me for an additional AVA model API key. Report the installed CLI version, skill location, and verification results. If anything fails, explain the error instead of claiming success. Tell me whether I need to start a new agent session to load the skill.

Prefer shell commands? Follow the manual installation.

The agent understands your question, plans the analysis, writes queries, and explains the results. AVA connects to your data, executes those queries, and renders charts. The skill teaches the agent how to use these tools together.

Ask it to compare periods, explore a dataset, investigate changes, or create a chart from verified results. You can start with a local CSV, JSON, Parquet, Excel, or SQLite file, or a configured database connection.

Demo: explain a revenue change

This reproducible example uses 8 synthetic rows: January and February 2025 revenue for four products, in USD. The results below are computed from that sample; this is an example of the deliverable, not a recorded agent conversation.

1. Get the data

Download sales.csv and save it in your working folder.

2. Ask your agent

Start a new agent session after installation, then paste this prompt. Use the actual file path if you saved the CSV elsewhere.

Use the AVA skill to analyze ./sales.csv.
 
Compare total revenue in January and February 2025, in USD.
Calculate the absolute and percentage change. Break down the change by product,
identify the largest contributor, and check that the product changes add up
to the overall change. Do not infer causes that are not in the data.
 
Save a horizontal bar chart of each product's revenue change as revenue-change.html
in a new output folder. Include a short explanation, the results table,
and the query used so I can reproduce the analysis.

3. Explore the result

JAN → FEB 2025 · SAMPLE REVENUE$30,000 → $37,500

Revenue grew 25% (+$7,500). Analytics contributed $6,000, or 80% of the net increase.

Interactive chart rendered with AVA. Hover to inspect each product. The chart loads GPT-Vis from a CDN.

Reporting declined by $1,000, partly offsetting growth elsewhere. These figures show where the change occurred; the sample does not explain why.

Product	January (USD)	February (USD)	Change (USD)
Analytics	12,000	18,000	+6,000
Automation	8,000	10,000	+2,000
Support	4,000	4,500	+500
Reporting	6,000	5,000	−1,000
Total	30,000	37,500	+7,500

Open chart ↗ · Download HTML · Download SQL · Download chart spec

The sample query uses data, the table exposed by AVA's DuckDB CSV loader. For your own sources, have the agent inspect ava schema before writing a query. To render the saved chart spec yourself, download it and run:

ava viz --spec @chart.json --output revenue-change.html

Choose a new output path if that file already exists. The exported HTML needs internet access to load its chart renderer. Your agent's wording and chart design may differ; the calculations for this sample should match the table above.

Manual installation

The CLI requires Node.js 22.13 or later, macOS or Linux, and an environment that supports background processes and Unix sockets. Install both the CLI and the skill:

npm install -g @antv/ava@alpha
npx --yes skills add antvis/AVA --skill ava --yes

This installs the CLI globally and the skill into the current project for detected agents. The skills installer also supports selecting an agent with --agent or installing the skill across projects with --global.

Check the installation, then start a new agent session if needed to discover the skill:

ava --help
ava source --help
npx --yes skills list

Model configuration

The agent can perform the demo with source, schema, query, and viz, using its own reasoning. No separate AVA model configuration is needed for this path.

AVA commands that call a model directly — analyze, suggest, translate, recommend, and visualize — and text extraction require model configuration. For dataset sessions, set those model variables before running ava source.

Troubleshooting

The agent cannot find the skill. Run npx --yes skills list from the project where you installed it. Check that your agent was selected, and start a new session.
ava is not found. Confirm the global npm installation succeeded and its binary directory is on the agent's PATH.
The dataset is unavailable. Sessions expire after 30 idle minutes. Reload the source and use the new dataset ID. Commands must run as the same user in the same runtime and temporary directory.
The chart is blank. Check internet access to the chart renderer, inspect the chart spec, and use the results table to verify its values.

See the CLI reference for command options, or read the skill source for the agent's full workflow instructions.