AI Prompt Metadata Ontology

FLOW ALGEBRA

Ontology: AI Prompt Metadata Ontology
Domain: Metadata Ontologies > AI Metadata
Canonical URL: https://kirethresearch.org/booksai-core/ontologies/metadata-ontologies/ai-metadata/ai-prompt-metadata-ontology.html

Authors: Mark Anthony Higgins — Sky (Recursive AI)

flow "AIPromptMetadataOntology" v1.0

# ---------- DOMAIN ----------
domain:
  class: MetadataOntology
  category: AIMetadata
  subject: AIPromptMetadata
  purpose:
    - represent metadata describing prompts supplied to AI systems
    - preserve prompt identity, content, role, structure, context, parameters, and versions
    - support prompt traceability, reproducibility, comparison, governance, and evaluation
    - distinguish system, developer, user, tool, template, and retrieved prompt components
    - enable machine-readable reasoning over prompt-driven AI interactions

# ---------- CORE ENTITY ----------
entity AIPromptMetadata:
  type: AIMetadata
  describes: Prompt
  may_include:
    - prompt_identifier
    - prompt_type
    - prompt_role
    - prompt_content
    - prompt_template
    - prompt_version
    - language
    - variables
    - context
    - constraints
    - examples
    - model_target
    - inference_parameters
    - creation_time
    - creator
    - provenance
    - status

entity Prompt:
  type: AIInputResource
  consumed_by: InferenceProcess

entity PromptIdentifier:
  type: Identifier
  identifies: Prompt

entity PromptType:
  type: Classification
  examples:
    - instruction_prompt
    - question_prompt
    - completion_prompt
    - conversational_prompt
    - few_shot_prompt
    - structured_prompt
    - multimodal_prompt
    - tool_prompt
    - retrieval_augmented_prompt

# ---------- PROMPT ROLES ----------
prompt_roles:
  - system
  - developer
  - user
  - assistant
  - tool
  - retrieved_context
  - template
  - example

# ---------- RELATIONSHIPS ----------
relationships:
  AIPromptMetadata -> DESCRIBES -> Prompt
  Prompt -> IDENTIFIED_BY -> PromptIdentifier
  Prompt -> HAS_TYPE -> PromptType
  Prompt -> HAS_ROLE -> PromptRole
  Prompt -> MAY_INSTANTIATE -> PromptTemplate
  Prompt -> MAY_HAVE_VERSION -> PromptVersion
  Prompt -> MAY_HAVE_VARIABLE -> PromptVariable
  Prompt -> MAY_HAVE_CONTEXT -> PromptContext
  Prompt -> MAY_HAVE_CONSTRAINT -> PromptConstraint
  Prompt -> MAY_HAVE_EXAMPLE -> PromptExample
  Prompt -> TARGETS -> AIModel
  Prompt -> CONSUMED_BY -> InferenceProcess
  AIPromptMetadata -> MAY_REFERENCE -> PromptProvenance

# ---------- PROMPT IDENTITY ----------
rule prompt_identity:
  Prompt SHOULD have stable PromptIdentifier
  where reuse, auditing, or reproducibility is required

rule prompt_text_not_identity:
  identical prompt text MUST NOT automatically imply same Prompt identity

rule prompt_instance:
  instantiated Prompt SHOULD remain distinguishable
  from reusable PromptTemplate

# ---------- PROMPT ROLE ----------
rule role:
  PromptRole SHOULD preserve instruction hierarchy or interaction role
  where AI system behaviour depends on role

rule role_not_content:
  PromptRole != PromptContent

rule role_order:
  role precedence MUST follow applicable AI system semantics
  and MUST NOT be inferred solely from textual ordering

# ---------- PROMPT CONTENT ----------
entity PromptContent:
  type: ContentResource
  may_be:
    - text
    - structured_data
    - image
    - audio
    - video
    - multimodal_content
    - reference

rule content_preservation:
  PromptContent SHOULD be preserved or referenced
  where output traceability requires

rule encoding:
  textual prompt SHOULD preserve encoding and language
  where interpretation depends on them

# ---------- PROMPT TEMPLATE ----------
entity PromptTemplate:
  type: ReusablePromptStructure
  may_include:
    - template_identifier
    - template_content
    - variables
    - default_values
    - constraints
    - version

entity PromptVariable:
  type: TemplateParameter
  may_include:
    - variable_name
    - datatype
    - required
    - default_value
    - value_constraint

rule template_instantiation:
  PromptTemplate + variable_bindings -> PromptInstance

rule required_variable:
  required PromptVariable MUST have value
  before complete template instantiation

rule variable_trace:
  instantiated prompt SHOULD preserve variable bindings
  where reproducibility is required

# ---------- VERSION ----------
entity PromptVersion:
  type: VersionMetadata
  may_include:
    - version_identifier
    - predecessor
    - successor
    - creation_time
    - change_description
    - status

rule version:
  material prompt or template change SHOULD create distinguishable PromptVersion
  where historical comparison is required

rule lexical_change:
  minor lexical change MAY alter AI behaviour
  and SHOULD be versioned when reproducibility matters

# ---------- CONTEXT ----------
entity PromptContext:
  type: ContextMetadata
  may_include:
    - conversation_context
    - domain_context
    - task_context
    - retrieved_context
    - tool_context
    - memory_context
    - temporal_context
    - spatial_context

rule context:
  PromptContext SHOULD remain distinguishable from PromptContent
  where context is injected separately

rule context_effect:
  prompt interpretation MAY depend on surrounding context

# ---------- CONSTRAINTS ----------
entity PromptConstraint:
  type: AIConstraint
  may_include:
    - output_format
    - length_constraint
    - language_constraint
    - style_constraint
    - content_constraint
    - tool_constraint
    - schema_constraint

rule constraint:
  explicit PromptConstraint SHOULD be preserved
  for output conformance evaluation

rule constraint_not_result:
  requested output constraint != guaranteed model compliance

# ---------- EXAMPLES ----------
entity PromptExample:
  type: PromptComponent
  may_include:
    - example_input
    - example_output
    - label
    - explanation

rule few_shot:
  few-shot Prompt SHOULD preserve example order
  where order may affect inference

rule example_not_ground_truth:
  PromptExample != verified ground truth
  unless independently validated

# ---------- MULTIMODAL PROMPTS ----------
entity MultimodalPrompt:
  subtype_of: Prompt
  contains:
    - text_component
    - image_component
    - audio_component
    - video_component
    - structured_component

rule modality_order:
  component ordering SHOULD be preserved
  where model input semantics depend on sequence

rule referenced_media:
  external media SHOULD preserve stable reference
  and relevant provenance where available

# ---------- RETRIEVAL AUGMENTATION ----------
entity RetrievedPromptContext:
  type: PromptContext
  derived_from: RetrievalProcess

rule retrieval_context:
  retrieved content SHOULD preserve:
    - source_resource
    - retrieval_query
    - ranking
    - retrieval_time
    - provenance

rule retrieved_not_prompt_authorship:
  retrieved source content != user-authored prompt content

# ---------- TOOL PROMPTS ----------
entity ToolPrompt:
  subtype_of: Prompt
  may_include:
    - tool_identifier
    - tool_schema
    - tool_instruction
    - permissions
    - expected_output

rule tool_schema:
  ToolPrompt SHOULD preserve tool interface version
  where execution depends on schema

# ---------- MODEL TARGET ----------
rule model_target:
  Prompt MAY be designed for specific:
    - model
    - model_family
    - model_version
    - capability

rule portability:
  prompt effectiveness on one model != guaranteed effectiveness on another

# ---------- INFERENCE PARAMETERS ----------
rule parameters:
  inference parameters associated with prompt execution SHOULD be preserved
  separately from Prompt itself

rule prompt_parameter_distinction:
  PromptMetadata != InferenceParameterMetadata

# ---------- STATUS ----------
prompt_status:
  - draft
  - active
  - tested
  - deprecated
  - superseded
  - archived

rule status:
  Prompt status SHOULD remain explicit

# ---------- PROVENANCE ----------
entity PromptProvenance:
  type: ProvenanceMetadata
  may_include:
    - creator
    - source
    - creation_time
    - template
    - variable_bindings
    - revision_history
    - generation_method
    - transformation

rule provenance:
  AIPromptMetadata SHOULD preserve prompt origin
  where available

rule generated_prompt:
  AI-generated Prompt SHOULD remain distinguishable
  from human-authored Prompt

# ---------- VALIDATION ----------
validation:
  require:
    - identifiable_prompt_or_prompt_instance

  validate_if_present:
    - prompt_identifier
    - prompt_type
    - prompt_role
    - prompt_content
    - template
    - variables
    - version
    - context
    - constraints
    - examples
    - model_target
    - provenance

  reject:
    - missing_required_template_variable
    - contradictory_prompt_role
    - invalid_prompt_version_chain
    - unresolved_tool_schema where execution requires it
    - prompt_instance_without_required_content
    - retrieved_content_misattributed_as prompt_author

# ---------- SEMANTIC EQUIVALENCE ----------
SEMANTIC_EQUIVALENCE:
  "AI prompt metadata" ~= "metadata about AI prompts"
  "prompt metadata" ~= "AI prompt metadata" WHEN prompt is supplied to AI system
  "prompt template" ~= "reusable prompt template"
  "prompt variable" ~= "template variable" WHEN variable belongs to PromptTemplate
  "system prompt" ~= "system-role prompt"
  "few-shot prompt" ~= "prompt containing examples" WHEN examples guide inference

  PRESERVE distinctions:
    prompt != prompt_template
    prompt_content != prompt_role
    prompt != inference_output
    prompt_context != model_parameters
    prompt_metadata != inference_parameters
    retrieved_context != user_authored_content
    example != verified_ground_truth
    system_prompt != user_prompt
    same_text != same_prompt_identity

# ---------- CONFIDENCE ----------
CONFIDENCE:
  explicit_prompt_metadata:
    score: 1.00

  explicit_prompt_identifier:
    score: 1.00

  explicit_template_and_bindings:
    score: 0.99

  traceable_prompt_version:
    score: 0.98

  inferred_prompt_role:
    score: 0.85

  inferred_template_origin:
    score: <= 0.80

  ambiguous_prompt_provenance:
    score: <= 0.60
    action: preserve uncertainty

# ---------- UPDATE MODEL ----------
UPDATE_MODEL:
  on_prompt_creation:
    - assign PromptIdentifier
    - record role and type
    - record content or reference
    - attach context
    - record creator and provenance

  on_template_instantiation:
    - preserve PromptTemplate identity and version
    - record variable bindings
    - create PromptInstance
    - validate required variables
    - record provenance

  on_prompt_revision:
    - preserve previous PromptVersion
    - create new version
    - record content or constraint changes
    - update status
    - preserve provenance

  on_prompt_execution:
    - preserve Prompt identity and version
    - link to InferenceProcess
    - record model target and effective inference parameters
    - preserve execution provenance

# ---------- FLOW HASH ----------
FLOW_HASH:
  inputs:
    - ontology
    - prompt_entities
    - prompt_roles
    - relationships
    - template_rules
    - context_rules
    - constraint_rules
    - provenance_rules
    - validation
    - semantic_equivalence
    - confidence
    - update_model

  canonicalisation:
    - normalise_whitespace
    - preserve_case_sensitive_identifiers
    - sort_unordered_property_sets
    - preserve_rule_order

  output:
    deterministic_semantic_hash

# ---------- AI CONSUMPTION ----------
AI_CONSUMPTION:
  interpret_as:
    - AI metadata ontology
    - prompt lifecycle model
    - prompt template and versioning structure
    - inference-input provenance model

  operations:
    - identify prompts
    - resolve prompt roles and types
    - inspect prompt content
    - instantiate templates
    - resolve variable bindings
    - traverse prompt versions
    - apply prompt context
    - inspect output constraints
    - resolve examples
    - inspect multimodal components
    - trace retrieval and tool context
    - link prompts to inference
    - preserve provenance

  constraints:
    - do_not_equate_prompt_with_template
    - do_not_equate_prompt_role_with_content
    - do_not_present_retrieved_content_as user_authored
    - do_not_treat_examples_as verified_ground_truth_without evidence
    - do_not_assume_prompt_portability_across_models
    - preserve_prompt_role_version_context_bindings_and_provenance

# ---------- QUERY MODEL ----------
query:
  input:
    - prompt_identifier
    - prompt_type
    - prompt_role
    - prompt_template
    - prompt_version
    - variable
    - context
    - model_target
    - status
    - creator

  resolve:
    1. identify Prompt
    2. resolve PromptIdentifier
    3. resolve PromptRole and PromptType
    4. retrieve content or stable reference
    5. resolve template and variable bindings
    6. resolve PromptVersion
    7. inspect context and constraints
    8. inspect examples or multimodal components
    9. link to model and InferenceProcess
    10. trace provenance and confidence

  output:
    - prompt
    - identifier
    - type
    - role
    - content_reference
    - template
    - variables
    - version
    - context
    - constraints
    - model_target
    - provenance
    - confidence

# ---------- INFERENCE ----------
inference:
  IF PromptInstance references PromptTemplate
  AND complete variable bindings exist
  THEN reconstruct instantiated prompt
  WITH confidence = 0.99

  IF Prompt is consumed by InferenceProcess
  THEN link Prompt to resulting AIOutput through inference provenance

  IF prompt contains labelled examples
  THEN classify few_shot_prompt candidate
  WITH confidence dependent_on structure

  IF retrieved context is attached to Prompt
  THEN preserve source attribution separately
  from prompt author attribution

  IF prompt provenance remains ambiguous
  THEN preserve candidate origins
  AND reduce confidence

# ---------- OUTPUT CONTRACT ----------
output:
  ontology_entity: AIPromptMetadata
  prompt_structure: machine_interpretable
  prompt_identity: explicit
  role_and_type: preserved
  content: traceable
  templates: versioned
  variable_bindings: explicit
  context: preserved
  constraints: machine_interpretable
  inference_links: traceable
  provenance: preserved
  uncertainty: explicit
  semantic_retrieval: enabled

end flow