Skip to main content

Quality Standard

Every skill ships proven, or it doesn't ship.

Nexus Certified means the skill you're installing has been tested to work on the platform you're using, in the way that platform expects, before it ever reaches the catalog.

What it means for you

You shouldn't have to wonder whether an AI skill is a genuine tool or a text file someone wrote in five minutes. Certification answers that question before you buy.

It works the first time you install it

Every skill is validated against the actual platform contract — not a generic template. Triggers fire correctly, output formats match what the platform expects, and tool calls resolve on the first run.

It's built for your platform, not copy-pasted across them

Each platform adapter is written and certified independently. A Claude Code skill uses Claude Code conventions. A Codex skill uses Codex conventions. No lowest-common-denominator compromises.

You get research-grade outputs, not prompt demos

We reject generic boilerplate, placeholder processes, and thin content. Every skill has a concrete output template, specific constraints, and enough depth to produce real deliverables — not "here are some thoughts."

Nothing ships without passing everything

There is no partial certification. A skill either passes every check on every platform, or it does not enter the catalog. The badge is binary.

Built for each platform

Claude Code, Codex, and OpenClaw each have different mechanics for loading, triggering, and executing skills. A generic skill that ignores those differences breaks at runtime. We certify each platform adapter independently.

🟠

Claude Code

Each skill follows the SKILL.md contract — correct frontmatter, semantic trigger descriptions, and explicit tool references — so Claude picks it up automatically the moment you install it. No configuration, no guessing.

🟢

Codex

Codex skills are structured instructions-first with explicit completion signals and bounded scope. They execute cleanly in Codex runs without runaway behavior, ambiguous output, or silent failures.

🔵

OpenClaw

OpenClaw skills ship with validated manifests and imperative playbooks that reference real tools by name. The platform loads them at startup and routes calls correctly — because the skill was built for how OpenClaw actually works.

The standard

Every skill-platform combination runs through a multi-layer audit before it enters the catalog. Structural integrity, platform compliance, and content quality — each layer is a gate, and any failure blocks the release.

Structure

Files exist, formats parse, sections are present, content meets minimum depth. No stubs, no placeholders, no skeleton files.

Compliance

Each adapter meets the target platform's contract — correct trigger mechanisms, valid tool declarations, proper completion signals.

Quality

No generic boilerplate, no template processes, no vague instructions. Concrete output formats, specific constraints, real substance.

96 checks across 3 platforms. Every skill, every release. No exceptions.

105

certified skill-platform combinations

96

checks per skill

0

ship without passing

Every skill in the catalog carries this standard.

105 certified combinations. 96 checks. Zero exceptions.

See every certified skill →