Quickstart

Install

One command installs the CLI with every Python feature. uv provisions Python 3.13 automatically:

$
uv tool install --python 3.13 "skillevaluator[all] @ git+https://github.com/NVIDIA/SkillEvaluator.git"

Confirm the install:

$
skillevaluator --version

You should see skillevaluator, version 0.1.0 (or newer).

Run the keyless quality check

Point SkillEvaluator at any skill directory (a folder with a SKILL.md):

$
skillevaluator quality-check ./my-skill

You get a deterministic quality score from 0–100 and an A–F grade; the check passes at 70 (a C) or above by default.

Run the complete offline Tier 1 validation

validate runs the full static and security check suite. The [all] install above includes the Python scanners, but complete evidence also requires the separate Semgrep, SkillSpector, and Gitleaks executables. Install all three, and pass --no-dedup to skip the Tier 2 pass that needs an embeddings key:

$
brew install semgrep gitleaks   # macOS
$
uv tool install git+https://github.com/NVIDIA/SkillSpector.git
$
skillevaluator validate ./my-skill --no-dedup

On other platforms, follow Installation: System tools for the Semgrep, SkillSpector, and Gitleaks setup. The terminal shows a compact pipeline view, and HTML and JSON reports land in reports/ — pass -r cli explicitly for terminal-only output, or --verbose for the full per-check detail stream. The command exits 0 only when every check passes and every required scanner returns complete evidence. A failed check or missing scanner evidence leaves the result INCOMPLETE and exits 1. See Gate Your CI for the full recipe.

Add an LLM provider

LLM-backed features — rubric judging, deeper security analysis, and Tier 2 deduplication — use whichever provider you configure. A free NVIDIA API Catalog key covers both LLM judging and embeddings with one variable:

$
export SKILL_EVAL_LLM_PROVIDER=nv_build
$
export NVIDIA_API_KEY='nvapi-...'

Check for redundancy

With a provider configured, find repeated guidance inside your skill, or compare every skill in a collection for overlap:

$
skillevaluator context-optimization-check ./my-skill   # inside one skill
$
skillevaluator similarity-check ./skills               # across a collection

Cross-skill matches are classified by similarity, from EXACT_DUPLICATE down to LOOSELY_RELATED; intra-skill duplicate findings include a suggested fix. See Tier 2: Deduplication.