Strategies

A strategy is a pure (level) -> config mapping implementing AdaptationStrategy. Resolve them directly (strategy.resolve(level)), through the engine (engine.resolve(strategy)), or in React via useStrategy. Each strategy also implements describe(level), returning a human-readable summary.

The package ships seven built-ins. Six are documented on this page:

Export	Strategy `name`	Config type	Module
`uiVisibilityStrategy`	`ui-visibility`	`UIVisibilityConfig`	core
`notificationStrategy`	`notifications`	`NotificationConfig`	core
`taskComplexityStrategy`	`task-complexity`	`TaskComplexityConfig`	core
`interactionForgivenessStrategy`	`interaction-forgiveness`	`InteractionForgivenessConfig`	core
`deferralStrategy`	`deferral`	`DeferralConfig`	core (see Deferral)
`autonomyStrategy`	`autonomy`	`AutonomyConfig`	core

The seventh, demandAdmissionStrategy, is documented with the inbound-demand policy it belongs to. createPresenceStrategy is a factory rather than a fixed strategy, and lives with Presence.

All resolved configs are frozen. All resolve implementations throw for invalid levels (they validate through getEnergyLevel).

uiVisibilityStrategy

const uiVisibilityStrategy: AdaptationStrategy<UIVisibilityConfig>

Which chrome elements are shown and how prominent they are.

UIVisibilityConfig

Prop

Type

Values per level

Level	sidebar	tabBar	statusBar	toolbar	chromeOpacity	chromeOpacityHover	contentMaxWidth	contentFontScale	readOnlyCursor
`100`	true	true	true	true	1	1	`none`	1	false
`75`	true	true	true	true	0.7	1	`none`	1	false
`50`	true	true	true	true	0.4	1	`90ch`	1	false
`25`	false	false	false	true	0.1	1	`80ch`	1.05	false
`0`	false	false	false	false	0.05	0.8	`75ch`	1.1	true

The same values are mirrored by applyEnergyLevel as CSS custom properties and by the reference stylesheet.

notificationStrategy

const notificationStrategy: AdaptationStrategy<NotificationConfig>

Which notification channels are allowed and how aggressively intents are filtered. Enforced at runtime by the notification gate.

NotificationConfig

Prop

Type

Values per level

Level	allowVisual	allowSound	allowVibration	batchInterval	priorityThreshold
`100`	true	true	true	0	`all`
`75`	true	true	false	0	`all`
`50`	true	false	false	300000 (5 min)	`high`
`25`	true	false	false	900000 (15 min)	`critical`
`0`	false	false	false	0	`none`

taskComplexityStrategy

const taskComplexityStrategy: AdaptationStrategy<TaskComplexityConfig>

Which tasks to surface and whether to nudge breaks.

TaskComplexityConfig

Prop

Type

Values per level

Level	maxComplexity	suggestBreaks	breakIntervalMinutes
`100`	`complex`	false	0
`75`	`moderate`	false	0
`50`	`routine`	true	45
`25`	`simple`	true	25
`0`	`consumption`	false	0

Rest (0) is already a break: prompting someone at 0 to take a break from resting is noise, so break suggestions are disabled entirely.

interactionForgivenessStrategy

const interactionForgivenessStrategy: AdaptationStrategy<InteractionForgivenessConfig>

How much room the interface gives the user to notice and reverse mistakes. Lower energy means slower error detection, so forgiveness scales inversely with capacity: longer undo windows, confirmation on destructive actions, more frequent autosave.

InteractionForgivenessConfig

Prop

Type

Values per level

Level	undoWindowMs	confirmDestructive	autosaveIntervalMs
`100`	5000	false	60000
`75`	8000	false	45000
`50`	10000	true	30000
`25`	15000	true	20000
`0`	20000	true	15000

deferralStrategy

const deferralStrategy: AdaptationStrategy<DeferralConfig>

Orders deferral presets so the default "not now" suggestion matches current capacity. Documented with its config and per-level values on the Deferral page.

autonomyStrategy

const autonomyStrategy: AdaptationStrategy<AutonomyConfig>

How much latitude automation has to act for the user without asking.

The mirror of interactionForgivenessStrategy: forgiveness protects against the user's mistakes at low energy, autonomy against the agent's. The system acts on the user's behalf precisely when they are least able to supervise it, so the worst day is the wrong day for it to improvise.

What narrows as energy falls is discretion, not action. At rest the threshold is 1, admitting only certainty - rule-based decisions, never a judgment call - and one unattended step. Automation may still take a single, certain, template-only action, which is exactly the shape of an out-of-office reply; it may not chain steps or compose wording.

AutonomyConfig

Prop

Type

Values per level

Level	confidenceThreshold	allowGeneratedContent	maxUnattendedSteps
`100`	0.6	true	8
`75`	0.7	true	5
`50`	0.8	true	3
`25`	0.9	false	1
`0`	1	false	1

const { confidenceThreshold, allowGeneratedContent, maxUnattendedSteps } =
  engine.resolve(autonomyStrategy)

if (classification.confidence < confidenceThreshold) return askTheUser()

Useful to any consumer with agentic surfaces, with or without demand admission.

Custom strategies

Any object satisfying AdaptationStrategy<TConfig> works with engine.resolve and useStrategy. For presence-shaped strategies, use createPresenceStrategy.

Strategies

On this page