Hallmark's Six Universal Disciplines: AI Content Quality Guardrails Explained

Hallmark applies six mandatory disciplines across every verb (default, audit, redesign, study, etc.) to eliminate AI-generated artifacts and enforce production-ready output standards. These disciplines are defined in skills/hallmark/SKILL.md and evaluated by an internal slop-test before any content is emitted.

Whether you're running the default verb to generate landing pages or the audit verb to review existing code, Hallmark enforces the same quality baseline. The codebase in Nutlope/hallmark implements these as non-negotiable guardrails rather than optional guidelines.

Pre-Emit Self-Critique: The Six-Axis Quality Gate

Before returning any artifact, Hallmark scores it on six evaluation axes: Philosophy, Hierarchy, Execution, Specificity, Restraint, and Variety. If any score falls below 3, the system forces a revision.

This discipline guarantees strategic and stylistic rigor regardless of output type. The scoring logic can be implemented as follows:

function preEmitCritique(artifact) {
  const scores = {
    philosophy: 5,
    hierarchy: 4,
    execution: 5,
    specificity: 4,
    restraint: 5,
    variety: 5,
  };
  const minScore = Math.min(...Object.values(scores));
  if (minScore < 3) {
    return { status: 'REJECT', revise: true, lowestAxis: getLowestAxis(scores) };
  }
  return { status: 'PASS', artifact };
}

The six-axis framework prevents shallow or mechanically-generated output from reaching production, as documented at SKILL.md L46.

Honest Copy: Zero Fabrication Policy

Hallmark never invents metrics, testimonials, or logos. All quantitative claims must use real numbers, explicit placeholders, or be omitted entirely.

This discipline directly targets a notorious AI failure mode: the confident generation of plausible-sounding but entirely fabricated statistics. The enforcement is strict—no exceptions for "realistic-looking" fake data, as specified at SKILL.md L48.

Locked Tokens: Design System Determinism

Every color and font-family must reference a named design token. Inline values like color: #3b82f6 or font-family: 'Inter' are prohibited.

This creates a single source of truth and makes the design system portable across projects. Token enforcement ensures that:

• Theme changes propagate through token redefinition
• No orphaned values exist in generated output
• Design systems remain maintainable at scale

The locked tokens rule is defined at SKILL.md L50.

Re-Drawn Chrome Forbidden: Authentic UI Only

Hallmark must not fabricate browser chrome, phone frames, IDE interfaces, or device bezels. Only real screenshots or clean content without decorative framing are permitted.

This eliminates the "fake UI chrome" anti-pattern that signals low-effort mock-ups rather than real product work. The restriction applies to all visual output regardless of verb, as documented at SKILL.md L52.

Mobile Responsiveness: Four-Breakpoint Verification

Every emitted page must render without horizontal scroll at 320px, 375px, 414px, and 768px breakpoints. Spacing, typography, and layout must remain correct across all four widths.

This sets a non-negotiable floor for production code. The mobile-first requirement is not aspirational—it is mechanically enforced before output approval, per SKILL.md L54.

Typography Purity: No Italic Headers

Headings and display type remain roman (upright). Emphasis is conveyed through weight, color, or underline—never through italics.

This removes a classic AI tell (italicized headings) and maintains clean typographic language. The prohibition applies to all heading levels and display text, as specified at SKILL.md L56.

Summary

• Pre-emit self-critique enforces six-axis quality scoring with mandatory revision on failure
• Honest copy eliminates fabricated metrics and testimonials entirely
• Locked tokens require all visual values to reference named design system tokens
• Re-drawn chrome forbidden bans fabricated browser/device frames
• Mobile responsiveness mandates verification at 320/375/414/768px breakpoints
• Typography purity prohibits italic headings in favor of weight and color emphasis

These six universal disciplines are defined in skills/hallmark/SKILL.md under the heading "Disciplines that hold across every verb" and evaluated by Hallmark's internal slop-test before any output is emitted.

Frequently Asked Questions

Where are the six universal disciplines defined in the Hallmark codebase?

The six disciplines are documented in skills/hallmark/SKILL.md at lines 46-56, under the section "Disciplines that hold across every verb." This file serves as the canonical specification for all quality guardrails applied regardless of which verb is executed.

How does Hallmark enforce these disciplines before output is returned?

An internal slop-test evaluates every artifact against the six disciplines before emission. The pre-emit self-critique specifically uses a six-axis scoring system (Philosophy, Hierarchy, Execution, Specificity, Restraint, Variety) with a minimum threshold of 3/5 on each axis, forcing automatic revision on any failure.

Do these disciplines apply to all Hallmark verbs equally?

Yes. The documentation explicitly states these disciplines hold across every verb—including default, audit, redesign, and study. No verb receives exemptions from the six-discipline framework, ensuring consistent output quality regardless of use case.

What happens if a design uses inline color values instead of tokens?

The artifact fails the locked tokens discipline. Hallmark requires every color and font-family to reference a named design token; inline values like hex codes are prohibited. This failure would trigger revision through the slop-test mechanism before any output is returned to the user.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}