Skip to content
Back to skills

Data And Replay Buffers

ASecurity

"Inspect Diffusion Policy dataset interfaces, ReplayBuffer stores,

  • 247 stars
  • 0 votes
  • 0 copies
  • 0 views
  • Added September 8, 2026
businesspythonapi

Works with

  • api

Security analysis

A100/100

Pro scans all 5 files and shows the line behind each finding

Scanned September 8, 2026

npx -y skills add VectorSpaceLab/AREX-Skill --skill data-and-replay-buffers --agent claude-code

Installs into .claude/skills of the current project.

Are you the author of Data And Replay Buffers?

Add the live security badge to your README. It updates with every re-scan.

Security grade badge for Data And Replay Buffers
[![Security: A — Skills Directory](https://www.skillsdirectory.com/api/skills/vectorspacelab-data-and-replay-buffers/badge)](https://www.skillsdirectory.com/skills/vectorspacelab-data-and-replay-buffers)

More formats (shields.io, HTML) on the badges page. Keep it an A: scan every change in CI with Pro.

Download with Pro
SKILL.md
---
name: data-and-replay-buffers
description: "Inspect Diffusion Policy dataset interfaces, ReplayBuffer stores,
  SequenceSampler horizons and padding, and normalizer contracts."
metadata:
  disco-role: operating
disable-model-invocation: true
license: MIT
---

# data-and-replay-buffers

Use this sub-skill when you need to understand or validate Diffusion Policy data stores, dataset adapters, sequence sampling, and normalization without touching training loops or policy internals.

## Use for
- zarr or zip ReplayBuffer inspection
- episode counts, `episode_ends`, shapes, dtypes, chunking, and length mismatches
- low-dimensional versus image dataset samples and `shape_meta` routing
- `SequenceSampler` horizon, padding, `episode_mask`, and `key_first_k` behavior
- normalizer selection, key lookup, and dataset conversion routes

## Route elsewhere
- training, evaluation, Ray multiruns, and metrics aggregation -> `training-and-evaluation`
- policy inference, checkpoint logic, model families, and losses -> `policies-and-models`
- live robot capture, RealSense, UR5, SpaceMouse, and safety-gated execution -> `real-robot-operations`

## Fast validation
1. Run `python scripts/inspect_replay_buffer.py --path <replay-buffer> --max-keys 12` from this sub-skill directory to inspect a zarr directory or zip store without modifying it.
2. Add `--json` when another tool should consume the summary.
3. Use the checklist in `references/troubleshooting.md` when episode lengths, keys, or shapes are inconsistent.

## What to read next
- `references/data-and-replay-buffers.md`
- `references/api-reference.md`
- `references/troubleshooting.md`

## Bundled helper
- `scripts/inspect_replay_buffer.py` — read-only inspector for ReplayBuffer zarr directories and zip stores.

## First checks when data looks wrong
- Compare `meta/episode_ends[-1]` to every `data/*` leading dimension.
- Confirm the sample shape matches the dataset family: flat low-dim tensor versus dict of image and state keys.
- Confirm the normalizer keys match the returned sample keys exactly.
- Treat repeated boundary frames as expected padding, not corruption.

Files in this skill

  • SKILL.md2.1 KB
  • references/api-reference.md7.4 KB
  • references/data-and-replay-buffers.md8.3 KB
  • references/troubleshooting.md6.1 KB
  • scripts/inspect_replay_buffer.py8.1 KB

Attribution

Is this your skill, or is something wrong with this listing? Request removal or report an issue. Author removals are honored within 72 hours.

Comments

Loading comments…

Newsletter

Get the weekly list of safe new skills, and the ones we pulled.