Core Concepts

Snapshot Modes

Use action/extraction/full snapshot modes correctly for interaction and data extraction.

Snapshot pipeline (prepareSnapshot):

Mark interactive/hidden/scrollable elements at runtime
Serialize page HTML
Clean with mode-specific cleaner
Assign counters to retained elements and sync the same counters to live DOM nodes

Cleaner modes:

All modes return compact, LLM-friendly HTML strings.

Counter semantics:

c values are assigned fresh on every snapshot pass.
The snapshot HTML and live DOM are synchronized to the same c values.
Counter resolution reads live DOM by c only (no snapshot session binding).
Action lookup uses strict unique match:
- no match -> target not found
- multiple matches -> target ambiguous
Extraction lookup is tolerant:
- no match -> field value is null
- multiple matches -> extraction fails as ambiguous
Boundary wrappers like os-iframe-root and os-shadow-root may be unnumbered.
Inaccessible iframes and closed shadow roots are not counter-addressable.

Mode	Description
`action`	Balanced context for action planning (default)
`extraction`	Richer content for data extraction
`clickable`	Clickable elements only
`scrollable`	Scrollable containers only
`full`	Broad HTML with scripts/styles/noise removed

Covered Features