Agent skills

Standr.

Professional judgment for AI agents

AI agents generate correct output. The problem is what they don’t generate: consistent naming, navigable architecture, tests that buy confidence. Standr gives your agent the domain judgment to close that gap.

standr/refactor
without standr with standr
def apply_coupon(subscription: Subscription, coupon_code: str) -> dict:
    # Fetch coupon from database
    data = db.get_coupon(coupon_code)

    # expires_at is stored in UTC — compare against utcnow()
    if data is not None and data.expires_at > datetime.utcnow():
        val = subscription.amount * data.discount_pct
        subscription.apply_discount(
            discount=val,
            promo_code=coupon_code
        )
        db.save(subscription)
        result = {"ok": True, "discount": val}
        return result

    return {"ok": False, "discount": 0}
def apply_coupon(subscription: Subscription, coupon_code: str) -> dict:
    coupon = db.get_coupon(coupon_code)

    # expires_at is stored in UTC — compare against utcnow()
    if not coupon or coupon.expires_at <= datetime.utcnow():
        return {"ok": False, "discount": 0}

    discount = subscription.amount * coupon.discount_pct
    subscription.apply_discount(
        discount=discount,
        coupon_code=coupon_code,
    )

    db.save(subscription)

    return {"ok": True, "discount": discount}

Works with

Claude Code OpenAI Codex CLI Gemini CLI OpenCode Pi Cursor GitHub Copilot Antigravity Kiro

AI output has a smell

You know it when you see it. Five endpoints that each invented their own conventions. An architecture that’s technically layered but structurally flat. A test suite that goes green without protecting a single invariant.

The output is correct. It just doesn’t look like someone cared.

Skills, not prompt packs

Prompt pack

A static text file. Front-loads instructions and hopes the model follows them. Quality varies with every run. Nothing adapts to your codebase.

standr skill

An executable workflow with judgment encoded. Reads your code, understands context, chains decisions, and makes targeted, defensible changes.

Install

$ npx skills add https://gitlab.com/standr.dev/standr

The best AI output is the kind nobody recognizes as AI.