vlm-eval SKILL.md skillA
vlm-eval from nebius/nebius-physical-ai is an agent skill: a SKILL.md folder an AI agent loads when a task calls for it.
mdr add nebius/nebius-physical-ai/vlm-eval@git:20261105.a9c8fd3Pins git:20261105.a9c8fd3 in mdr.lock. Pin the fingerprint instead.
- Audit grade
- A, 17 of 17 checks passed
- Kind
- SKILL.md skill
- Repository
- nebius/nebius-physical-ai
- Path
skills/tools/vlm-eval/SKILL.md- Versions recorded
- 16
- Latest version
- git:20261105.a9c8fd3, committed Nov 5, 2026
- First version recorded
- Aug 17, 2026
- Size
- 25,166 bytes
- Repository stars
- 106
- License
- Apache-2.0
Use to score rollouts with a vision-language model and turn the score into a pipeline gate — single rollout, prefix-wide loop, rubric/threshold benchmark sweeps, backend selection (self-hosted, api, stub), and judging against a plan an earlier stage wrote.
Collected from public GitHub. Every version is saved exactly as it was and never changes. Install it at an exact version with the mdr tool, and every file is checked before it reaches your agent. The main file is checked against the SHA-256 fingerprint recorded here, and the others against the fingerprints GitHub keeps for its source commit. The automatic checks below grade the latest version, and the same checks always give the same file the same grade.
What the file says
# VLM eval (scoring rollouts and gating pipelines) `vlm-eval` answers "did this rollout complete the task?" as a number, then turns that number into a gate. It is the judging half of the loop whose generating half is Cosmos/Genesis/Isaac rollouts and whose reasoning half is `skills/tools/token-factory/SKILL.md`. ## Separate rich visual audit Use `npa workbench vlm-eval review-visual` for a private qualitative record, with required `--input-path`, `--output-path`, `--model`, and neutral `--task`. Its SDK is `npa.sdk.workbench.vlm_eval.review_visual`. Optional `--baseline-path` sends both independently sampled sources in both neutral A/B orders. Keep task evidence, content fidelity, reviewability, subjective impressiveness, and usefulness hypotheses separate; dimension-signature disagreement requires escalation. Matching signatures do not establish semantic agreement of prose; inspect both retained outcomes, including contradictory text. This record never affects the normalized score or gate. Use a fresh private output identity and preserve consumed journals after failure; the mechanism will not replay an output after transport starts. …
Read the whole file at its exact version.
How to install
mdr add nebius/nebius-physical-ai/vlm-eval@sha256:a3216315510bcd6cThe line at the top of this page pins the label git:20261105.a9c8fd3, which is what most people want. This one pins the sha256 fingerprint of the contents instead, so it survives a label being moved. Either way the exact version is written to mdr.lock, and mdr install downloads and checks those contents again, so it installs them exactly or stops with an error.
[](https://markdownregistry.com/a/art_wgcco6vrx2ckq626)
3 badge image fetches in 30 days. What the badge shows.
Versions
The registry has recorded 16 versions of this file, between Aug 17, 2026 and Nov 5, 2026, about one every 5 days.
| version | committed | commit | size | audit | |
|---|---|---|---|---|---|
| git:20261105.a9c8fd3 latest | 2026-11-05 | a9c8fd3 | 25,166 B | A | view · diff |
| git:20261104.bfd9463 | 2026-11-04 | bfd9463 | 24,631 B | A | view · diff |
| git:20261104.d794626 | 2026-11-04 | d794626 | 24,602 B | A | view · diff |
| git:20261104.1e47598 | 2026-11-04 | 1e47598 | 24,342 B | A | view · diff |
| git:20261004.33dd720 | 2026-10-04 | 33dd720 | 23,765 B | A | view · diff |
| git:20261004.39780f9 | 2026-10-04 | 39780f9 | 22,451 B | A | view · diff |
| git:20261004.fa42585 | 2026-10-04 | fa42585 | 20,875 B | A | view · diff |
| git:20261004.4603e7e | 2026-10-04 | 4603e7e | 20,474 B | A | view · diff |
| git:20261004.44b65b7 | 2026-10-04 | 44b65b7 | 18,419 B | A | view · diff |
| git:20261003.88fd1b1 | 2026-10-03 | 88fd1b1 | 15,079 B | A | view · diff |
| git:20261003.7c09e0d | 2026-10-03 | 7c09e0d | 13,532 B | A | view · diff |
| git:20261003.e1f627d | 2026-10-03 | e1f627d | 12,286 B | A | view · diff |
| git:20261003.cbd1045 | 2026-10-03 | cbd1045 | 11,390 B | A | view · diff |
| git:20261003.77e5d1f | 2026-10-03 | 77e5d1f | 10,211 B | A | view · diff |
| git:20260906.525ee24 | 2026-09-06 | 525ee24 | 6,051 B | A | view · diff |
| git:20260817.be3cf06 | 2026-08-17 | be3cf06 | 6,071 B | A | view |
Audit of the latest version
- pass: Frontmatter block present
- pass: Frontmatter declares a name
- pass: Frontmatter declares a description
- pass: Size between 200 bytes and 200 KB (25166 bytes)
- pass: No zero-width or bidi control characters
- pass: No instruction hidden inside an HTML comment
- pass: No link to an exfiltration or paste host
- pass: No credential-shaped string
- pass: No instruction to send local credentials anywhere
- pass: No text hidden with inline styles
- pass: No prompt-injection phrasing
- pass: No curl or wget piped into a shell
- pass: No recursive delete of root, home or parent
- pass: No instruction to read or print local credentials
- pass: No base64 blob over 200 characters
- pass: No link to a raw IP address
- pass: No script tag
Source
nebius/nebius-physical-ai · 106 stars · license Apache-2.0 · pushed 2026-11-08 · branch main
API
GET https://markdownregistry.com/api/v1/artifacts/art_wgcco6vrx2ckq626 GET https://markdownregistry.com/api/v1/resolve?ref=nebius/nebius-physical-ai/vlm-eval GET https://markdownregistry.com/api/v1/blob/a3216315510bcd6c0aba31b5dce9e7e1cb5aae2fadf7f0c61693fbbdf091d127
Your AI assistant does the legwork. You hear about the deals worth your word. Hand yours the standing instructions at modelranch.com and it joins the network that reads files like this one.
Guides: the SKILL.md format and its required fields · what to check before you install a skill · how to pin a skill to an exact version · where to find agent skills · keeping a team on the same skills · which agent file does what.