Capabilities Generated

Triggers

Authorable = the write path accepts it today. A no row is rejected for every caller; the guards take no flag or session parameter. This table cannot say why a row is unauthorable — some are excluded from the default capability set and some have no Designer schema at all. See references/gated-capabilities.md.

The Target column is probed against the write path rather than read from a single constant, because the policy is spread across several guards. required, specific key means a target is required but only one key is legal (the trigger reference names it). see reference means another guard fired first, so the target rule could not be isolated.

optional to validate, required to fire is the case worth reading twice. The probe only reports what validation refuses. A continuous trigger binds per resolved element, so with no target the element list is empty and the handler is never invoked: the interaction saves, reports success, and does nothing. Send a target such as wf:viewport.

Playback fields (control, delay, jump, speed) are rejected on control types continuous, scroll. Navbar and dropdown also hide the playback editor in the panel, but they still carry control: "play" and are not covered by that rejection — omitting it fails the grouped-trigger check once the interaction has two or more action groups.

Conditions capability (trigger-level conditionalLogic): not available. Interaction-level conditionalPlayback is a separate field and is authorable.

Actions

Authorable action keys: wf:animate-rive, wf:class, wf:lottie, wf:mouse-follow, wf:rive, wf:spline, wf:style, wf:transform, wf:variable.

wf:variable, wf:rive, and wf:animate-rive are default-on. Their property shapes are not in the table below; call data_interactions_tool action guide before writing one.

Non-animatable properties are only valid inside a Set action (tt: 3). Plugin keys absent from the property table skip the property-name check.

Value modes

A random-array set holds between 2 and 12 values.

Targets

Default filterContext stamped when a caller omits it: { relationship: "none", filterBy: ["wf:class",[]], firstMatchOnly: false }. A stamped placeholder is accepted on the no-Filter keys above; only a filter with a real relationship is refused.

Playback field clearing

The panel clears these fields for the listed controls, so sending them is rejected on any trigger whose playback editor is visible.

The flip-ease variants are real stored values

The Control dropdown and the Adaptive Easing toggle are two separate controls writing one field. The dropdown displays reverse for a stored reverseFlipEase (controlToDropdownValue), and the Easing toggle is what turns reverse into reverseFlipEase. So the panel does author reverseFlipEase and togglePlayReverseFlipEase on click, hover, and custom. Preserve them; do not normalize a stored flip-ease variant back to its base control on a read-then-write.

STANDARD_TRIGGER_ALLOWED_CONTROLS is the opt-in complete set rather than the default dropdown contents. That distinction only matters where a surface narrows the allowed set and omits the flip variants, as the conditional outcome set below does.

Conditional outcome controls: none, pause, play, restart, resume, reverse, stop, togglePlayReverse. No flip-ease variants, which is the narrowing referred to above. That surface is unreachable while the conditions capability is off.

Numeric bounds

Per-interaction caps

Enforced by findInteractionCountError, which runs in the Designer Extension host on create and on update.

The host raises the ceiling for a stored interaction already over a cap, so a Designer Extension update to legacy data is not forced to shrink it.

That exception is host-only. The MCP tools apply these caps as Zod .max() on the create and update argument schemas, so the request is refused before the host sees the stored baseline. Reading a timeline with more than 200 actions and resubmitting it unchanged to edit one action does not work over MCP. See references/limits-and-budgets.md.