Interface: ComposedIntent
Defined in: aiui-lowering-pipeline/src/types.ts:648
Properties
components
components:
LocatedComponent[]
Defined in: aiui-lowering-pipeline/src/types.ts:661
corrections
corrections:
object[]
Defined in: aiui-lowering-pipeline/src/types.ts:653
applied
applied:
boolean
instruction
instruction:
string
original
original:
string
patch?
optionalpatch?:string
scope?
optionalscope?:object
The chunk window the fix was scoped to (see the correction event).
scope.fromLine
fromLine:
number
scope.toLine
toLine:
number
items
items:
ComposedItem[]
Defined in: aiui-lowering-pipeline/src/types.ts:652
Chronological interleave of text runs, shots, and selections (app + code).
prompt
prompt:
string
Defined in: aiui-lowering-pipeline/src/types.ts:673
The lowered body: prose with each screenshot inlined at its position as a [screenshot located at <path>] bracket line plus, when elements were located, a <screenshot-metadata> XML block — path (relativized against ComposeOptions.cwd when given), located elements, and their cell frontier, all in the text where the image belongs. This replaced the Option-C {shot_n} token + meta-map scheme: the indirection cost a hint line and a metadata block the agent had to correlate, for structure nothing downstream actually consumed (meta only ever became text attributes on the rendered channel tag).
spans
spans:
PromptSpan[]
Defined in: aiui-lowering-pipeline/src/types.ts:682
Offset-annotated structure over prompt: shots, on-screen and code selections, navigation/tab boundaries, and — once the channel wraps the body — the context preamble. Consumers (the trace hero) render the raw prompt and overlay hover-previews / source hyperlinks / a de-emphasized preamble from these instead of re-parsing the string. Purely additive: the prompt text is byte-identical whether or not anyone reads spans.
transcript
transcript:
string
Defined in: aiui-lowering-pipeline/src/types.ts:650
Transcript with replace-policy corrections applied.