Function Reference Model Input/Output Data
SubAIMs JSON Metadata Profiles

1. Function

The Prompt Creation AIM (PGM-PRC) assembles, aligns, and exposes perceptual and user‑related semantics to BKN. It structures perception outputs and user context into a prompt suitable for reasoning, and acts as the authoritative access point for assembled perceptual and user semantics, without performing perception or domain reasoning itself.

PRC produces

  • Raw Goal Expression, a structured, pre‑semantic representation of the user’s current goal, inferred from available perceptual and interaction evidence and not independently invented by PRC. by assembling and structuring Perception‑derived and interaction‑derived outputs, including:
    • ASR outputs (ASD1),
    • VSR outputs (VSD1),
    • User Entity State snapshots.
  • Context Perceptual Semantics by assembling and aligning:
    • Multimodal alignment cues across audio, visual, and interaction modalities.
    • Referent bindings and disambiguation structures.
    • Spatial, temporal, and relational indicators.
    • User‑related interpretations derived from perception (e.g. attention, speaking, engagement),
    • Uncertainty and alternative interpretations present in perception outputs.

PRC does not enhance or reinterpret perception outputs; it composes and indexes semantics already produced by perception AIMs.

PRC constructs the PRCPrompt, which includes:

  • The Raw Goal Expression.
  • The Context Perceptual Semantics.
  • References to relevant perceptual and user‑related descriptors stored in the A‑User Store,
  • A  SessionID enabling BKN to retrieve Context snapshots (VSD0, ASD0, UES0), accumulated perceptual descriptors, A-User Entity State records, and Interaction History.

PRCPrompt construction avoids duplication of perceptual data and relies on references for retrieval.

PRC initiates and manages Multi Call Protocol (MCP) sessions as required by BKN, thus enabling

  • Structured and typed semantic queries.
  • Structured and typed responses.
  • Session‑level continuity across multi‑stage clarification exchanges.

PRC acts as the perceptual‑semantic access point during MCP interactions, similar to DAC’s role for domain semantics.

During clarification loops initiated by BKN, PRC:

  • provides additional perceptual and user‑related semantics derived from existing descriptors,
  • refines multimodal alignment and referent disambiguation,
  • responds to BKN semantic queries concerning ASD, VSD, UES snapshots, and their relationships.

PRC may request additional perception outputs when required, but does not generate new perceptual facts.

In conclusion, PRC acts as:

  • An Interpretation AIM, structuring perceptual evidence into goal‑relevant representations, and
  • A Perceptual Semantic Provider AIM, supplying BKN with assembled, aligned, and queriable perceptual and user semantics.

2. Reference Model

Figure 1 gives Reference Model of Prompt Creation (PGM-PRC).

Figure – The Reference Model of Prompt Creation (PGM-PRC)

3. Input/Output Data

Table 1 – Input/Output Data of PGM-PRC

Input Description
Context A structured and time-stamped snapshot representing the initial understanding that the A-User achieves of the environment and of the User posture.
Audio Scene Descriptors A Data Type that conveys spatially grounded semantic audio data from the Spatial Reasoning AIM that enables the Prompt Construction AIM to generate context-aware and referentially precise prompts.
Visual Scene Descriptors A Data Type that conveys spatially grounded semantic visual data from the Spatial Reasoning AIM that enables the Prompt Construction AIM to generate context-aware and referentially precise prompts.
Prompt Creation Directive Trigger to initiates prompt generation or refinement from PGM-AUC.
Output Description
PC-Input A machine‑interpretable MCP‑structured object containing PRC’s interpretation of Context, User Intent, referents (objects, people, or scene elements referred to by User’s vague expressions), User State, and Audio and Visual Scene Descriptors. Sent as input to PGM-BKN.
Prompt Creation Status
Structured MCP metadata expressing PRC readiness, alignment, and processing state. Sent to and used by PGM‑AUC for orchestration and control.
A structured MCP data object enumerating the MCP tools exposed by Prompt Creation, each defined by its name, purpose, input schema, and output schema. This manifest enables PGM‑BKN – and when appropriate, Domain Access – to discover and invoke PRC’s internal computational functions through the MCP Server.

4. SubAIMs

Table 2 describes the informative SubAIMs that have been identified for PGM-PRC.

Table 2 – Informative SubAIMs

SubAIM Function Inputs Outputs To
CIP – Context Integration & Parsing Assemble multimodal context and Entity State for prompt; select exact referents/relations in prompt scope. ASD1; VSD1; ASD3; VSD3; Entity State Item Zone Map; Validation Flags CVV
CVV – Context & Validation Verifier (Prompt‑Scoped) Final verification of consistency wrt input-text-referenced items/relations; apply policy gates; check zone feasibility. Item Zone Map; ASD1; VSD1; ASD3; VSD3; Entity State; Validation Flags Validation Status; Zone Feasibility Tag; Scene Consistency Flag BSL
BSL – Behaviour Selection & Logic Retrieve and rank candidate behaviours based on validated context and Entity State. Entity State; Validation Status; Scene Consistency Flag Behaviour Candidates List EFR
EFR – Execution Feasibility & Recovery Assess feasibility and produce fallback or adjusted prompt when needed. Behaviour Candidates List; Validation Status; Zone Feasibility Tag Execution Feasibility Status; Fallback Action PCP
PCP – Prompt Composition & Planning Prepare structured PC-Prompt Plan (JSON) integrating validated context, constraints, and behaviours. Execution Feasibility Status; Behaviour Candidates List; Scene/Entity Tags PC-Prompt Plan PFT
PFT – Prompt Formatter (new) Convert PC-Prompt Plan (JSON) into natural language PC-Prompt for BKN. PC-Prompt Plan PC-Prompt BKN
LTM – Lifecycle Trace Metadata Record provenance and traceability for audit/lifecycle management. ASD1 ID; ASD3 ID; VSD1 ID; VSD3 ID; Execution Feasibility Status Trace Log AUC
  • Table 3 defines the terms used.
  • Table 3 – Terminology used in this AIM
Term Definition
ASD1 / VSD1 Audio Scene Descriptor (ASD1) and Visual Scene Descriptor (VSD1) generated by Context Capture. These represent the initial scene understanding before refinement.
ASD3 / VSD3 Audio Scene Descriptor (ASD3) and Visual Scene Descriptor (VSD3) produced by ASR/VSR after their final interaction cycle with DAC. These are the most refined descriptors available to PRC.
Entity State A structured representation of the current state of relevant entities (e.g., position, orientation, status flags) in the scene. Replaces the older concept of “User State.”
Item Zone Map A mapping of referenced items (from the input text) to their spatial zones and relationships in the scene, derived from ASD1/VSD1 and ASD3/VSD3. Used to ground prompt content.
Validation Flags Indicators generated by CIP to highlight potential ambiguities or risks (e.g., unresolved referents, missing zone info) for CVV to resolve.
Validation Status The result of CVV’s checks (OK / WARN / FAIL) including reasons and confidence scores for multimodal consistency and policy compliance.
Zone Feasibility Tag A tag indicating whether the spatial zone referenced in the prompt is physically/logically feasible for the intended action.
Scene Consistency Flag A graded indicator confirming that the scene context assumed by the prompt matches the actual scene descriptors.
Behaviour Candidates List A ranked list of possible behaviours/actions retrieved by BSL based on validated context and Entity State.
Execution Feasibility Status A status object indicating whether the selected behaviour(s) can be executed under current constraints (e.g., zone reachability, operational limits).
Fallback Action An alternative behaviour or adjusted prompt generated by EFR when execution feasibility fails.
PC-Prompt Plan A structured JSON plan integrating validated context, constraints, and behaviours for natural language generation. Used internally by PRC to produce the PC-Prompt.
PC-Prompt The natural-language prompt assembled by PRC and sent to BKN (LLM).
Trace Log Provenance and lifecycle metadata recorded by LTM for audit and orchestration purposes.
  • Table 4  maps PRC Inputs/Outputs to Unified Messages.
  • Table 4 — PRC Inputs/Outputs mapped to Unified Messages
PRC Data Name Role Origin / Destination Unified Schema Mapping
Context Input From Context Capture (CXC) Consumed by PRC as contextual basis for prompt generation; carried in Context; referenced via Envelope.CorrelationId; MUST include Trace.Origin and Trace.Timestamp.
Prompt Creation Directive Input From A‑User Control (AUC) Directive → TargetAIM=PRC; includes generation parameters (style, constraints, modality); correlation maintained via Envelope.CorrelationId.
Entity State (optional) Input From Context Capture (CXC) If used by PRC: carried in Context; referenced for personalization or posture-aware prompt adaptation.
Prompt Object Output To BKN Status → Result (generated prompt text or multimodal structure); maintain Envelope.CorrelationId; MUST include Trace.Origin and Trace.Timestamp.
Prompt Creation Status Output To A‑User Control (AUC) Status → State/Progress/Summary/Result; includes confidence, generation steps, and any fallback logic; MUST include Trace.Origin and Trace.Timestamp.