vla-evaluation-harness: Skill for Claude Code

.claude/skills/run-evaluation/SKILL.md

run-evaluation is a skill for Claude Code from allenai/vla-evaluation-harness. It costs 87 tokens per session (2,006 once invoked), scanned A, original, Apache-2.0.

A workflow for measuring a vision-language-action (VLA) model against a simulated robotics benchmark, using separate model-serving and benchmark processes.

In plain words
What is it for?
It helps serve a model, launch benchmarks such as LIBERO or CALVIN in Docker, check required tools and hardware, and save evaluation results.
Why use it?
It organizes the setup needed to run model tests in simulation and checks that the model and benchmark use compatible action formats.

Skill for Claude Code

Written for Claude Code: installed under .claude/.

This is allenai/vla-evaluation-harness's own configuration. It tells Claude Code how to work on vla-evaluation-harness itself, so it is not a mod to install elsewhere. Copy it as a starting point and replace the rules that are about this project. Everything vla-evaluation-harness configures →

Needs its repository: it runs a file that does not travel with it, so clone the repository first. The line is ./docker/build.sh <benchmark_name> # e.g. ./docker/build.sh libero.

Reuse

Borrowing it

Nothing to install: this file belongs to allenai/vla-evaluation-harness. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.

Copy the file
curl -O https://raw.githubusercontent.com/allenai/vla-evaluation-harness/main/.claude/skills/run-evaluation/SKILL.md
Clone the repo
git clone --depth 1 https://github.com/allenai/vla-evaluation-harness

Made for: Claude Code.

Wrote this? Show the measurements

A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.

agentmods badge for run-evaluation

README.md
[![agentmods](https://agentmods.dev/badge/skills/allenai/vla-evaluation-harness/run-evaluation/github.svg)](https://agentmods.dev/skills/allenai/vla-evaluation-harness/run-evaluation)
Your own site
<a href="https://agentmods.dev/skills/allenai/vla-evaluation-harness/run-evaluation"><img src="https://agentmods.dev/badge/skills/allenai/vla-evaluation-harness/run-evaluation/github.svg" alt="Measured on agentmods" height="20"></a>

Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.

agentmods 80×15 button for run-evaluation

Your own site · 80×15
<a href="https://agentmods.dev/skills/allenai/vla-evaluation-harness/run-evaluation"><img src="https://agentmods.dev/badge/skills/allenai/vla-evaluation-harness/run-evaluation.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 87 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 2,006 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 1 finding. A grade says what 26 rules found in the file — not that it is safe. Third-party audits
  • NVIDIA SkillSpector warn 7 Sept 2026
SkillSpector: 2 findings, up to high

These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →

  • high Privilege Escalation · line 69
    Potential security issue detected. Manual review is recommended.
    Fix: Review the flagged content for security risks. Ensure no credentials, secrets, or sensitive data are exposed.
  • medium MCP Rug Pull · line 34
    Docker image references without a specific tag (:latest is implicit) or digest (@sha256:...) can be silently replaced by a malicious image.
    Fix: Pin the image: image:tag or image@sha256:abc123
How audits are shown
Origin original No closer match found in the catalogue.
Token cost

What it costs to keep this loaded

Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.

ModelPer sessionOnce invoked
Fable 5.1 $0.00087 $0.02006
Opus 5 $0.00044 $0.01003
Sonnet 5 $0.00017 $0.00401
Haiku 4.5 $0.00009 $0.00201

Measured 9d ago against content hash d38bf678edda, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-08, from the pricing page.

Security

Grade A, and why

run-evaluation scanned grade A with 1 finding against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 9d ago.

A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.

Makes network callslowCapability

Not a fault in itself. Listed so you know the mod talks to something, and to what.

Wait until `curl -fsS http://localhost:8000/health` returns HTTP 200 — the server only starts listening after `__init__` finishes loading weights, so this is the readiness signal.
.claude/skills/run-evaluation/SKILL.md · 179 lines

How it starts

The opening of the file, as written. The whole thing — 179 lines — stays where its author put it; the contents beside it link to each section on GitHub.

Run Evaluation

Evaluate a VLA model against a simulation benchmark. The harness decouples model serving (WebSocket server) from benchmark execution (Docker container), so they run as two separate processes.

1. Identify the config pair

Every evaluation needs two YAML configs:

  • Model server config (configs/model_servers/<model>.yaml) — defines script and args for the model server
  • Benchmark config (configs/<benchmark>.yaml) — defines docker.image, benchmarks entries, and output_dir

List available configs:

ls configs/model_servers/    # model servers
ls configs/*.yaml            # benchmarks

Not all model–benchmark pairs are compatible. The model server must produce actions in the format the benchmark expects (e.g. 7-DoF for LIBERO). Many model configs encode their target benchmark in the filename (e.g. oft_libero.yaml, xvla_calvin.yaml).

2. Check prerequisites

Requirement Check command Notes
uv which uv Runs model server in isolated env
Docker docker info Benchmarks run inside containers
GPU nvidia-smi Model inference + sim rendering
Disk space df -h Model weights (tens of GB) + Docker images (4–10 GB each)

Model weights download automatically on first vla-eval serve. Docker images are pulled on first vla-eval run (or pre-pull with docker pull <image>).

Docker image rebuild: Benchmark code runs inside the Docker image. If you (or someone else) changed benchmark source code in src/vla_eval/benchmarks/, the pre-built image is stale — you must rebuild before running:

./docker/build.sh <benchmark_name>   # e.g. ./docker/build.sh libero

Skip the rebuild only if using --dev mode, which bind-mounts local src/ into the container.

3. Run the evaluation (two terminals)

The model server and benchmark runner communicate over WebSocket and must run concurrently.

Terminal 1 — start the model server:

vla-eval serve -c configs/model_servers/<model>.yaml

Wait until curl -fsS http://localhost:8000/health returns HTTP 200 — the server only starts listening after __init__ finishes loading weights, so this is the readiness signal.

Read the full file on GitHub · 179 lines

Changes

What this file has done since we first saw it

Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.

  1. 9d ago First seen · 179 lines · 87 tokens per session scan A d38bf678edda

Subscribe to this mod's changes

run-evaluation is a skill published in the GitHub repository allenai/vla-evaluation-harness (591 stars, last pushed 7d ago), licensed Apache-2.0. It adds 87 tokens to every session and 2,006 once invoked, about $0.0004 per session on Opus 5. A static security scan graded it A with 1 finding (makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.