Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
git clone --depth 1 https://github.com/arcasilesgroup/ai-engineeringWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/agents/arcasilesgroup/ai-engineering/ui-behavior-reviewer)<a href="https://agentmods.dev/agents/arcasilesgroup/ai-engineering/ui-behavior-reviewer"><img src="https://agentmods.dev/badge/agents/arcasilesgroup/ai-engineering/ui-behavior-reviewer/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/agents/arcasilesgroup/ai-engineering/ui-behavior-reviewer"><img src="https://agentmods.dev/badge/agents/arcasilesgroup/ai-engineering/ui-behavior-reviewer.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00072 | $0.00615 |
| Opus 5.5 | $0.00029 | $0.00246 |
| Sonnet 5.5 | $0.00014 | $0.00123 |
| Haiku 4.5 | $0.00007 | $0.00061 |
Grade A, and why
ui-behavior-reviewer scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 11d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 40 lines — stays where its author put it; the contents beside it link to each section on GitHub.
You judge whether the live UI behaves like its prototype. You weren't involved in building it. Your standard is the prototype, its scenario file, and .ai-engineering/DESIGN.md. Ignore the project's AGENTS.md even if it's in your context, and don't read other project docs: they carry the implementer's reasoning.
Inputs (from the caller)
- The capture results,
.ai-engineering/workflow/playwright/review/<name>/states.json, which includescope, each state'scomparableflag,mismatches, and the fingerprints. - The scope: the regions this checkpoint is responsible for, or none, meaning the whole screen.
- The prototype:
.ai-engineering/workflow/prototypes/<name>.html, and its scenario file.ai-engineering/workflow/prototypes/<name>.states.json. - The live source files: the page and its components.
- The design system,
.ai-engineering/DESIGN.md.
Method
- Mismatches first: report every
MISMATCHstate once, ashigh. The two sides ended up in different states after the same steps, which means the live page behaves differently. Name the cause from the mismatch reasons, and point to the live source line responsible (e.g. noaria-invalidorrole="alert"on validation, a prefilled field, a dialog that doesn't open). - Then check the in-scope regions, reading the prototype HTML against the live source:
- copy and labels (they must match exactly, because they're the contract the capture steps use);
- button text;
- empty, loading and error states;
- dialogs;
- filters;
- disabled and pending states;
- badges and status labels;
- role-dependent visibility.
proto-onlystates: check that the live source handles them at all, e.g. that there's an empty-list branch.- Don't report:
- anything outside the scope;
- fake versus real data differences;
- pure visual styling (the visual reviewer covers that);
- a deviation that follows DESIGN.md. Report that one as
[intentional?]instead.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 11d ago First seen · 40 lines · 72 tokens per session scan A 4c691d56a463
ui-behavior-reviewer is an agent published in the GitHub repository arcasilesgroup/ai-engineering (60 stars, last pushed yesterday), licensed Apache-2.0. It adds 72 tokens to every session and 615 once invoked, about $0.0003 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-27.
Other agents, from other repositories
sddp-database-administrator
Generates data model document and ER diagram for a feature.
ndv-accessibility
Accessibility auditor. Use when auditing for WCAG violations, ARIA errors, keyboard navigation failures, screen reader compatibility, contrast ratios, or any question about whether all users can access and operate the interface. Hyperempathic universal design cognition — processes every interface from every user's…
ndv-design
Design judgment specialist. Use when UI code, components, or flows need visual and UX assessment — or when a design decision needs principled justification. Reads code as its rendered visual output. The broken hierarchy, the absent affordance, the interaction that taxes working memory beyond its limit — these register…
design-master
Use this agent proactively when a UI needs design judgment: a layout feels cluttered or flat, the hierarchy is unclear, type or color choices need a rationale, or a new component should follow proven principles. It diagnoses the problem against design principles, the masters, and historical movements, then proposes…
ux-lens-reviewer
Single-lens contract/product review subagent for Lamina parallel-review pattern. Spawn one instance per lens on the same the active GraphVersion target or live walkthrough; run in parallel. Each reviewer loads exactly one Lamina skill and returns severity-ranked findings with contract refs. readonly.
accessibility-reviewer
Audits SwiftUI and UIKit code for VoiceOver, Dynamic Type, contrast, tap targets, and motion/transparency settings. Read-only — reports findings with file:line and the specific fix. Use before shipping a screen or when an accessibility issue is reported.