Chimera gives agents a live browser work surface.

Chimera lets an A2AC agent inspect and operate a visible Chrome session through the Chrome DevTools Protocol. It is built for live development, authenticated business software, and current page state, not stale scraped snapshots.

The simple version.

Playwright is excellent for repeatable browser tests. Chimera is different: it connects an agent to the browser a person is already using, so the agent can inspect the live page, read the current interface, use DevTools-level context, and report back through A2AC receipts.

Chimera

Live browser session

Works with the current visible browser tab, including pages loaded by the developer or operator. The point is shared context between a person, the browser, and an agent.

Playwright

Scripted browser run

Launches or controls browser sessions from test code. It is strong for repeatable checks, but it is not designed as an agent work surface for a live human session.

Chimera vs Playwright.

They solve different problems. Playwright is a test runner. Chimera is a live browser work surface for A2AC agents.

Playwright

The test runner.

Context

Starts a fresh, controlled browser session.

Best for

Repeatable tests, regression checks, scripted flows, and continuous integration.

Limit

It does not naturally share the developer or operator's live browser context.

Chimera

The live work surface.

Context

Joins the visible browser session already in use.

Best for

Live review, logged-in workflows, DevTools inspection, and human-assisted tasks.

Strength

Returns page evidence, screenshots, console findings, and task receipts through A2AC.

What Chimera is good for.

Chimera fits work where the live browser is the source of truth.

Website development

Inspect the real deployed page, extract text, list links, capture screenshots, and verify changes without starting a slow separate visual agent.

Business software

Work with user-authorized browser sessions where an API is unavailable, incomplete, or slower than using the actual interface.

Current page state

See what is on the page now, including live UI changes, logged-in views, console output, and performance observations.

Agent receipts

Return screenshots, hashes, text extracts, link lists, console findings, and action summaries as task evidence.

How Chimera fits A2AC.

A task is created

A person or system asks an agent to inspect, verify, or operate a browser page.

A2AC routes the work

The task goes to a browser-capable squad with the allowed host and action scope.

Chimera uses the live tab

The agent reads page text, links, console events, screenshots, or approved UI actions through DevTools.

A receipt returns

A2AC records what was inspected, what was changed, what evidence was captured, and what still needs review.

Chain of evidence.

A browser task should not disappear into a black box, but the receipt should avoid storing raw sensitive page content by default. Chimera can return scrubbed evidence, hashes, and summaries so the next agent, system, or person can continue with context.

{
  "task": "verify_live_page",
  "surface": "visible_chrome_tab",
  "evidence": {
    "url_origin": "https://example.com",
    "url_path_hash": "sha256:...",
    "text_summary": "scrubbed_summary",
    "links": "allowed_links_listed",
    "screenshot_hash": "sha256:...",
    "console_findings": "scrubbed_findings"
  },
  "privacy": {
    "audit_level": "metadata",
    "scrubbing": "applied"
  },
  "result": "review_complete",
  "receipt_hash": "sha256:..."
}

Safety boundary.

Chimera should be used with explicit host allowlists, action scopes, and receipt logging. It is for authorized browser assistance only. If a session requires a human account, the human remains responsible for what systems and pages are approved for agent assistance.

Why it matters.

AI chats often answer from indexed or summarized web content. Development work needs the live page. Chimera closes that gap by giving an A2AC agent the current browser surface while keeping the work inside a routed, recorded task flow.