Module a11y 

Accessibility-tree snapshots with element refs.

This module is pure: it turns the JSON returned by CDP Accessibility.getFullAXTree into a compact, Playwright-style YAML listing that agents can read cheaply, and it hands out stable element refs ([ref=eN]) that the native input path (crate::session::input) resolves back to backendDOMNodeIds. Nothing here talks to a browser, so every rule is unit-tested against canned tree fixtures.

Why the AX tree rather than an injected JS walker: Chromium’s accessibility tree is already composed across shadow roots, carries the computed accessible name/role the same way Playwright reports them, and attaches a backendDOMNodeId per node — exactly the handle DOM.* and Input.* accept. A DOM walker would have to reimplement accessible-name computation and could not cross closed shadow roots.

Re-exports

pub use refs::RefEntry;

pub use refs::RefTable;

Modules

refs

Per-tab element reference table.

Structs

AxNode

One node of the accessibility tree, normalised from an AXNode.

AxTree

The whole tree, indexed by nodeId.

FindOptions

Options for find.

Match

One find hit.

Snapshot

Rendered snapshot.

SnapshotOptions

Rendering options for render_snapshot.

Constants

DEFAULT_FIND_LIMIT

DEFAULT_MAX_CHARS

Functions

document_token

Identity of the document the tree was taken from: the root (RootWebArea) node’s backendDOMNodeId, which is the Document node. It changes on every navigation, so it doubles as a staleness token for refs. None when the tree carries no backend ids (mock/partial).

find

Score every candidate node against a free-text query. Pure text matching — no model call. Returns at most opts.limit matches, best first, ties broken by document order.

is_interactive_role

Roles that mean “the agent can act on this”.

parse_full_ax_tree

Parse a Accessibility.getFullAXTree result ({ nodes: [...] }).

render_snapshot

Render the tree as indented YAML-ish lines, interning refs as it goes.