AgentGlass — Agent Readiness Console

SCANNER · live3 agents onlinequeue · 2 sitesregion · us-east-1

Live scan

Scan your site.

3 agents · 1 site · 24 seconds — Filter the event stream before your patch.

Maya Lin · Boltline · boltline.example

t+0.0sscripted replayno external services

t+0.0s / 24.0s

agentglass scan --agents=3 --duration=24s

waiting for first event…

Search

How AgentGlass actually tests your site

33 agentsClaude · OpenAI · Gemini
parallel sandboxes

↻Run the same production journeyidentical task, real DOM

100Score per agent0-100 readiness

✓Adversarial verifierchecks the patch set

PRPatches → PRsafe merge path

Each of the 3 agents independently navigates the same production journey in parallel sandboxes. Per eval cell, we replay 5 runs and average against ground-truth success criteria — not LLM-as-judge.

What's running under the hood

Multi-agent

3 browser agents run the same journey in parallel

Multi-step

Each agent navigates a multi-page production flow, not single-page actions

Evals

Every scan replays 5 runs per cell against ground-truth success criteria — not LLM-as-judge

Self-improvement loop

Failed patches feed back into the patch generator's training signal nightly

Sandboxed agents

Each agent runs in its own isolated Chromium container — no shared cookies, no cross-contamination

Realtime via Redis

Scan events stream to the dashboard via Redis pub/sub — sub-200ms latency

Containerized DB

Postgres + Redis run in docker-compose locally; same containers in production

Findings

Expandable failure cards.

Every finding expands inline with the before/after DOM snippets and a concrete fix summary. No modal, no hidden state.

Add explicit aria-label so agents can match intent without relying on inner text + class semantics.

Before

<button class="btn-primary cta-book-demo">Book a demo</button>

After

<button class="btn-primary cta-book-demo" aria-label="Book a demo with our sales team">Book a demo</button>

Promote to focusable group with role + label.

Before

<div class="slots" tabindex="-1">...</div>

After

<div class="slots" tabindex="0" role="radiogroup" aria-label="Available booking slots">...</div>

Add a visible close button; agents don't reliably send Esc.

Before

<div class="modal" role="dialog">...</div>

After

<div class="modal" role="dialog" aria-labelledby="modal-title"><button aria-label="Close dialog">×</button>...</div>

Replace mouseover dependency with click handler + expose ARIA state.

Before

<div class="size-selector" onmouseover="openMenu()">M</div>

After

<button class="size-selector" onclick="openMenu()" aria-expanded="false" aria-haspopup="listbox">M</button>

Use polite live region and persist until agent acknowledges so post-click verification succeeds.

Before

<div class="toast" role="alert">Added to cart</div>

After

<div class="toast" role="status" aria-live="polite" data-persist-until-ack="true">Added to cart</div>

Add an explicit close button — agents can't easily target backdrops.

Before

<div class="backdrop" onclick="closeCart()"></div>

After

<button class="close-cart" aria-label="Close cart" onclick="closeCart()">×</button>

Disambiguate via aria-label per tier.

Before

<button>Continue</button><button>Continue</button><button>Continue</button>

After

<button aria-label="Continue with Free plan">Continue</button> ... etc

Single click is the universal pattern; double-click is an agent dead zone.

Before

<button class="copy-key" ondblclick="copy()">Copy</button>

After

<button class="copy-key" onclick="copy()" aria-label="Copy API key to clipboard">Copy</button>

Render the nav at all viewports — agents often run in 1024×768 sandboxes.

Before

<aside class="docs-nav" data-hidden-below="1200"></aside>

After

<aside class="docs-nav" aria-label="Documentation navigation"></aside>  // remove viewport gate

Use semantic <button> + role=tab so agents and screen readers understand the affordance.

Before

<div class="tab" onclick="swap('py')">Python</div>

After

<button class="tab" role="tab" aria-selected="false" onclick="swap('py')">Python</button>

Promote to status so agents include it in their reasoning context.

Before

<div class="banner" role="presentation">Free shipping over $50</div>

After

<div class="banner" role="status" aria-label="Promotional offer">Free shipping over $50</div>

Use <th scope=row> so agents parse rate-limit values correctly.

Before

<tr><td>Requests/sec</td><td>100</td><td>1000</td><td>10000</td></tr>

After

<tr><th scope="row">Requests/sec</th><td>100</td><td>1000</td><td>10000</td></tr>

Patch queue

DOM diffs agents can navigate.

Verified patches — safe to merge. They passed our adversarial verifier with all 3 agents.

components/BookDemoCTA.tsx

critical verified

Booking CTA missing aria-label

components/BookDemoCTA.tsx

- <button className="btn-primary cta-book-demo">Book a demo</button>+ <button className="btn-primary cta-book-demo" aria-label="Book a demo with our sales team">Book a demo</button>

components/ui/Modal.tsx

medium needs review

Modal dismiss requires Esc — no close button — Verified-false — adversarial verifier flagged: dismiss button needs tab-trap fix in same component for full AA compliance

components/ui/Modal.tsx

+ <button onClick={onClose} aria-label="Close dialog" className="modal-close">×</button>

components/SizeSelector.tsx

critical verified

Size selector requires hover-then-click

components/SizeSelector.tsx

- <div className="size-selector" onMouseOver={openMenu}>{size}</div>+ <button className="size-selector" onClick={openMenu} aria-expanded={open} aria-haspopup="listbox">{size}</button>

pages/dashboard/api-keys.tsx

critical verified

API key copy button requires double-click

pages/dashboard/api-keys.tsx

- <button className="copy-key" onDoubleClick={copy}>Copy</button>+ <button className="copy-key" onClick={copy} aria-label="Copy API key to clipboard">Copy</button>

components/docs/CodeTabs.tsx

high verified

Code example tabs are <div> with click handler — no keyboard

components/docs/CodeTabs.tsx

- <div className="tab" onClick={() => setLang('py')}>Python</div>+ <button className="tab" role="tab" aria-selected={lang==='py'} onClick={() => setLang('py')}>Python</button>

agents.json

Generated site contract.

The base schema and shape come from seed data; persona switching updates the visible site identity and score using seeded persona values.

{
  "$schema": "https://agentglass.dev/schemas/agents.v1.json",
  "version": "1.0",
  "site": {
    "name": "Boltline",
    "url": "https://boltline.example",
    "description": "B2B SaaS — scheduling — Booking CTA has no aria-label"
  },
  "agents_welcome": true,
  "ax_score": 94,
  "ax_score_breakdown": {
    "labeling": 96,
    "keyboard": 92,
    "non_semantic_handlers": 95,
    "hover_dependencies": 89,
    "transient_state": 97,
    "responsive_blockers": 100,
    "ambiguous_targets": 91
  },
  "scan_history": [
    {
      "scanned_at": "2026-05-10T14:00:00Z",
      "score": 34
    },
    {
      "scanned_at": "2026-05-11T14:00:00Z",
      "score": 64
    },
    {
      "scanned_at": "2026-05-12T14:00:00Z",
      "score": 94
    }
  ],
  "supported_journeys": [
    {
      "name": "book-demo",
      "entrypoint": "/",
      "selector_hints": {
        "primary_cta": "[aria-label='Book a demo with our sales team']",
        "calendar_slots": "[role='radiogroup'][aria-label='Available booking slots']",
        "confirm": "button.confirm-booking"
      },
      "expected_completion_ms": 12000,
      "min_score_for_agent_traffic": 80
    },
    {
      "name": "view-pricing",
      "entrypoint": "/pricing",
      "selector_hints": {
        "tier_growth": "button[aria-label='Continue with Growth plan']",
        "tier_scale": "button[aria-label='Continue with Scale plan']"
      }
    }
  ],
  "rate_limits_for_agents": {
    "requests_per_minute": 60,
    "scan_concurrency": 3
  },
  "contact": {
    "owner": "maya.lin@boltline.example",
    "agentglass_dashboard": "https://agentglass.dev/orgs/boltline"
  }
}

Integration

How customers ship the outputs

AgentGlass does not stop at a score. It hands teams a manifest path, a patch path, or both — depending on what they want to put into production first.

URLCustomer pastes URLboltline.com
production target

3Spin up agents3 sandboxed Chromium agents
running in parallel

↻Run the journey“book a demo”
real DOM, fixed task

●Record failuresevery click, retry,
timeout, and miss

3Generate outputsAX score · DOM diffs
agents.json manifest

✓Customer integrateshost the manifest,
apply patches, or both

Two integration models, chosen per output: publish agents.json so agents know how to navigate, or apply DOM patches so the site itself becomes easier for agents to use.

Host the manifest

Download agents.json and host it at /.well-known/agents.json on your domain. Agents read it before navigating — like robots.txt for AI.

manifest check200 OK

curl https://customer.com/.well-known/agents.json
# Claude / GPT / Operator read this before navigating

Apply the patches

Three options: download the .diff and apply manually, enable GitHub auto-PR for one-click merges, or drop in the runtime snippet for zero-deploy fixes.

⬇manual .diff

🔀GitHub auto-PR

</>runtime snippet

30-cell scorecard

Tiny cells, visible numbers.

10sites
1task each
3agents
5runs averaged

boltline.example highlighted

82book-demo
claude-cu

47book-demo
openai-op

38book-demo
gemini

76add-to-cart
claude-cu

61add-to-cart
openai-op

44add-to-cart
gemini

91copy-api-key
claude-cu

58copy-api-key
openai-op

67copy-api-key
gemini

88create-doc
claude-cu

72create-doc
openai-op

51create-doc
gemini

79log-call
claude-cu

54log-call
openai-op

41log-call
gemini

71upload-episode
claude-cu

49upload-episode
openai-op

36upload-episode
gemini

94toggle-flag
claude-cu

81toggle-flag
openai-op

73toggle-flag
gemini

68export-figma
claude-cu

53export-figma
openai-op

39export-figma
gemini

85search-and-edit
claude-cu

67search-and-edit
openai-op

58search-and-edit
gemini

62update-card
claude-cu

41update-card
openai-op

29update-card
gemini

GREENAgent completed the journey (score ≥ 75)
AMBERPartial recovery (50–74)
REDFailed or timeout (< 50)

Decision picker

Which plan fits your team?

Four clicks produce a live recommendation from the seed-data decision logic.

Where does most of your traffic come from today?

Do you ship code on a weekly cadence or faster?

Have you instrumented bot vs. human traffic?

What's your team size for frontend changes?

Pricing

Pricing — Start free, scale on CI

All plans share the same scanner and adversarial verifier. The difference is volume, automation depth, and auto-PR access.

Free

$0 Forever, no card required

1 site · 10 scans/month · community support

Public AX score (badge on your homepage)
agents.json export (download as .json)
Manual patch download (.diff)
Single failure-mode lookup

Growth

$199 per month

10 sites · unlimited scans · email support

Everything in Free
CI mode — runs on every GitHub deploy
Patch auto-PR (requires repo permission greenlight)
PDF Agent Readiness Report (weekly + on-demand)
Slack alerts on AX score regressions
Custom URL list (10 sites)
Private agents.json hosting

Scale

Contact us Custom pricing

Unlimited sites · SLA · SSO · dedicated CSM

Everything in Growth
Private model routing (BYO LLM keys)
Custom AX score weighting (per-vertical tuning)
Multi-tenant workspaces (agencies)
On-prem deployment option
SOC 2 evidence pack
Quarterly business reviews

Find the DOM bugs AI agents trip on.

Scan your site.

How AgentGlass actually tests your site

What's running under the hood

Expandable failure cards.

DOM diffs agents can navigate.

components/BookDemoCTA.tsx

components/ui/Modal.tsx

components/SizeSelector.tsx

pages/dashboard/api-keys.tsx

components/docs/CodeTabs.tsx

Generated site contract.

How customers ship the outputs

Host the manifest

Apply the patches

Tiny cells, visible numbers.

Which plan fits your team?

Where does most of your traffic come from today?

Do you ship code on a weekly cadence or faster?

Have you instrumented bot vs. human traffic?

What's your team size for frontend changes?

Pricing — Start free, scale on CI

Free

Growth

Scale

Get the AgentGlass field guide