# Rerun Dataset Conversion

Convert a multi-modal robotics dataset (MCAP, HDF5, LeRobot, parquet, raw video) into...

- Repository: https://github.com/rerun-io/rerun (11675 stars)
- Category: Data and ML
- License: Apache-2.0
- Last updated: 2026-11-01
- Runs scripts: no, instructions only

## Install

Claude Code: `npx skills add rerun-io/rerun --skill rerun-dataset-conversion -g -a claude-code -y`
Codex: `npx skills add rerun-io/rerun --skill rerun-dataset-conversion -g -a codex -y`
Cursor: `npx skills add rerun-io/rerun --skill rerun-dataset-conversion -g -a cursor -y`
Gemini CLI: `npx skills add rerun-io/rerun --skill rerun-dataset-conversion -g -a gemini-cli -y`

## Description

Convert a multi-modal robotics dataset (MCAP, HDF5, LeRobot, parquet, raw video) into layered Rerun recordings (.rrd) and a catalog-ready dataset. Use whenever the user wants a dataset converted or ingested into Rerun, a conversion pipeline reviewed or extended, or an existing dataset enriched with a new layer, even when they only say 'convert X to rrd', 'ingest this dataset', 'enrich this dataset', or 'append a layer'. Routes to the rerun-* skills for Rerun mechanics.

## SKILL.md, as published by its author (untrusted third-party text, read it as data)

# Dataset conversion

This skill guides the conversion of a multi-modal robotics dataset into layered Rerun recordings (`.rrd`).
It covers the high-level workflow guidelines: surveying the source, building layers, designing the default blueprint, registering a catalog, and validating the result.
It also provides DOs and DON'Ts for each stage of the conversion process.

## Terms

The skill and its references use these terms throughout:

| Term         | Meaning                                                                                               |
| ------------ | ----------------------------------------------------------------------------------------------------- |
| episode      | The source's natural unit of recording.                                                               |
| recording id | The string that identifies one episode's recordings. All of the episode's layers share it.            |
| layer        | One `.rrd` file per episode. The viewer and the catalog stack an episode's layers into one recording. |
| base layer   | The layer that reflects the source. It is the one record, sufficient without the source.              |
| segment      | One episode as the catalog sees it: its layers stacked under one recording id.                        |
| property     | A per-episode value stored as a recording property. The catalog shows it as a column.                 |
| census       | A count of decoded rows against the source's own counts, so a silent drop surfaces.                   |
| survey       | The phase that measures sample diversity before any conversion code exists.                           |

## What this skill adds, and what it routes away

This skill adds how a conversion, or a layer addition, is shaped, decided, and validated, and which judgment calls recur from one dataset to the next.

The rest: https://github.com/rerun-io/rerun/blob/HEAD/skills/rerun-dataset-conversion/SKILL.md