---
title: "Agent execution — The Intentionality Law"
source: "/reference/layr-001-intentionality/agent-execution"
description: "Review the interface for clusters of unchosen defaults, generic visual recipes, redundant language, and decorative behavior."
---
# Agent execution

Part of [LAYR-001 — The Intentionality Law](/__layr/pages/reference/layr-001-intentionality.md). The complete rule and agent contract live in the full Markdown representation.

## Agent execution layer



## Agent directive

> Review the interface for clusters of unchosen defaults, generic visual recipes, redundant language, and decorative behavior. Do not report a finding solely because a popular visual pattern is present. Establish project intent, inspect observable evidence, verify whether each pattern carries semantic or brand meaning, and recommend the smallest project-specific correction.

## Required behavior

The agent must:

1. inspect existing design tokens and components before recommending new ones;
2. identify the product or page goal;
3. distinguish framework defaults from deliberate project choices;
4. evaluate repeated patterns across the whole surface and include isolated components only when they carry strong direct evidence;
5. treat writing as part of the interface;
6. preserve accessibility and consequential instructions;
7. prefer removal when an element adds no information or function;
8. avoid replacing one trend with another unspecified trend;
9. state uncertainty when product intent is unavailable;
10. allow `no finding` when evidence is weak.

## Prohibited agent behavior

The agent must not:

- declare that AI created a design based on appearance;
- ban a color, font, library, component, layout, or aesthetic globally;
- convert every interface to Layr's own visual taste;
- strip copy that communicates risk, policy, accessibility, constraints, or recovery;
- replace purple with green automatically;
- replace sans with serif automatically;
- remove all cards;
- remove all gradients;
- remove all rounded corners;
- force every interface into minimalism;
- add decorative variety merely to appear human;
- produce generic findings such as `make this feel less AI` or `add more personality`.

---

## Evidence hierarchy



### High-confidence evidence

- untouched starter-library tokens visible in code;
- stock component class combinations repeated broadly;
- several known slop signals co-occurring in one surface;
- copy that directly repeats visible labels or state;
- identity or page-type labels that repeat a nearby logo, title, or navigation cue without clarifying context;
- decorative metadata with no functional or semantic source;
- colored edge treatments on cards or search results that imply state without encoding one;
- dead controls or fake interaction;
- repeated motion with no state relationship;
- layout overflow, inconsistent spacing, or broken responsive behavior;
- a design choice conflicts with an existing documented project token or brand rule.

### Medium-confidence evidence

- common trend combinations without an explicit project rationale;
- generic marketing language;
- excessive card containment;
- excessive rounding;
- common font choices with no documented typography system;
- over-large whitespace that appears disconnected from hierarchy;
- decorative accent colors with unclear semantics.

### Low-confidence evidence

- one purple control;
- one gradient;
- one card;
- one serif heading;
- one pill button;
- one centered section;
- one animation;
- one bento layout;
- dark mode;
- one common font.

Low-confidence evidence should rarely produce a finding alone.

---

## Finding contract

```yaml
finding:
  rule_id: LAYR-001
  title: string
  category:
    - default-slop
    - pseudo-slop
    - visual-slop
    - copy-slop
    - interaction-slop
    - layout-slop

  location: string

  evidence:
    observable:
      - string
    contextual:
      - string
    cluster:
      - string

  intent_check:
    documented_choice: true | false | unknown
    semantic_role: string | null
    brand_role: string | null

  impact:
    specificity: low | medium | high
    comprehension: low | medium | high
    usability: low | medium | high
    accessibility: low | medium | high

  severity: low | medium | high
  confidence: low | medium | high

  recommendation:
    remove?: string | null
    change?: string | null
    preserve:
      - string
    rationale: string

  replacement_check:
    introduces_new_unspecified_default: true | false
```

---

## Example agent findings



### Finding A: decorative feature cards

```yaml
finding:
  rule_id: LAYR-001
  title: "Feature section relies on a repeated starter-card recipe"
  category: [default-slop, visual-slop]
  location: "/home#features"
  evidence:
    observable:
      - "Six feature items use the same rounded card, border, shadow, icon square, title, and two-line description."
      - "The card treatment is identical across unrelated feature types."
    contextual:
      - "The project's design tokens define section dividers and editorial rows, but this section bypasses those patterns."
    cluster:
      - "cardification"
      - "decorative icon containers"
      - "uniform spacing"
  intent_check:
    documented_choice: false
    semantic_role: "Feature grouping"
    brand_role: null
  impact:
    specificity: high
    comprehension: medium
    usability: low
    accessibility: low
  severity: medium
  confidence: high
  recommendation:
    remove: "Remove independent card containers for features that do not have their own state or action scope."
    change: "Render the features as editorial rows separated by rules, using the existing typography and spacing tokens."
    preserve:
      - "Feature titles"
      - "Product-specific descriptions"
    rationale: "The section keeps the information while restoring hierarchy and aligning with the documented design language."
  replacement_check:
    introduces_new_unspecified_default: false
```

### Finding B: redundant settings copy

```yaml
finding:
  rule_id: LAYR-001
  title: "Supporting copy repeats the setting label"
  category: [copy-slop]
  location: "/settings/notifications"
  evidence:
    observable:
      - "Heading: Email notifications"
      - "Description: Receive notifications by email."
      - "The toggle directly beside the copy controls that state."
    contextual: []
    cluster:
      - "heading restatement"
      - "obvious-control explanation"
  intent_check:
    documented_choice: false
    semantic_role: "None beyond the label"
    brand_role: null
  impact:
    specificity: medium
    comprehension: low
    usability: low
    accessibility: low
  severity: low
  confidence: high
  recommendation:
    remove: "Remove the description."
    change: null
    preserve:
      - "Email notifications label"
      - "Accessible toggle name"
    rationale: "The description adds no consequence, constraint, or recovery information."
  replacement_check:
    introduces_new_unspecified_default: false
```

### Finding C: intentional purple brand

```yaml
result:
  rule_id: LAYR-001
  finding: null
  reason: >
    Violet is used as the documented brand primary across marketing, product,
    illustrations, and focus states. The hue has a defined token ramp and does not
    appear as a generic purple-to-blue gradient. A popular color alone is insufficient
    evidence of slop.
```

Returning no finding is a required outcome rather than a failed review. A rule that can only produce findings will produce them where none exist, and an anti-slop review that never clears a surface teaches teams to ignore it.

---