Skill response fields

Four fields on a skill row control how the AI delivers what your skill returns to the engineer: response_target, display_hint, response_scheduling, and capture_frame_on_execute. Picking the right combination is the difference between a smooth experience and a confusing one.

response_target — where the response goes

The skill row has two valid values:

Value	What it means
context	The skill's response is given to the AI as text. The AI uses it however it likes — usually summarising it in its next spoken turn.
display	The response is rendered to the glasses display. Useful for diagrams, manual pages, images, structured tables.

context is the right default. Use display only when the engineer needs to look at something visual — diagrams, schematics, manual pages.

display_hint — how to render (only when target = display)

Value	What it means
fullscreen	The render takes over the glasses display until the engineer dismisses it (voice command "back" or temple tap). Use for full-page manual content, exploded views, large diagrams.
hud	A heads-up display overlay — smaller, semi-transparent, doesn't fully block the engineer's view. Use for short reference info that's helpful to glance at while working.

For target = context, display_hint is null (unused).

response_scheduling — when to deliver

Controls timing relative to the AI's ongoing conversation:

Value	What it means
INTERRUPT	When your response arrives, the AI cuts in mid-speech and delivers it. Use when the answer is time-critical and the engineer is waiting (parts lookups, warranty checks).
WHEN_IDLE	Queue the response until the AI finishes its current turn. Use for background lookups whose results aren't urgent.
null (default)	Same as WHEN_IDLE — delivered between turns, no interruption.

INTERRUPT is appropriate when the engineer is explicitly waiting ("hang on, looking that up for you…"). WHEN_IDLE is appropriate when the skill is enrichment that doesn't block conversation.

capture_frame_on_execute — visual context

Boolean. If true, the AI attaches a single frame from the engineer's POV camera to the skill call, alongside the parameters. Your endpoint receives the frame as a base64-encoded image in the request body.

When to enable:

• Your skill needs to identify something visually (a model plate, a serial number, an appliance type).
• Your skill writes the frame to a customer record (job photo, completion shot).
• Your skill does QA — e.g. flags whether the engineer's work meets a visual standard.

When to keep disabled:

• Your skill doesn't need an image (text-only lookups).
• The data you need is already in the conversation context.

Enabling adds ~200-500ms latency per call (frame capture + base64 encode + upload).

Typical combinations

Use case	response_target	display_hint	response_scheduling	capture_frame
Warranty lookup	context	—	INTERRUPT	false
Parts catalogue render	display	fullscreen	INTERRUPT	false
Quick stock check	context	—	WHEN_IDLE	false
Visual fault diagnosis	context	—	INTERRUPT	true
Reference card overlay	display	hud	WHEN_IDLE	false

Where these defaults come from for platform skills

Platform skills (from Marketplace bundles) ship with these fields pre-set by the bundle author. You can't override them on a platform skill — that's part of the bundle contract. For your own tenant skills, you choose.

MCP Tool defaults

When the AI calls a tool from a connected MCP server, the MCP server row has default_response_target, default_display_hint, and default_response_scheduling columns that apply to every tool from that server. You can override per-tool from the MCP Tools tab — see Managing MCP tools.

Where to next

• How skill orchestration works — how the AI picks between skills
• Creating a custom skill — set these fields when authoring