Knowledge Entity Representation

Overview

Use this skill when a digital property depends on entities and claims that systems or people may confuse. It creates a source-backed representation contract before downstream search, generative-visibility, structured-data, or agent-action work.

The skill is consultative. It may identify ambiguity, propose canonical descriptions, and recommend updates. It must not invent facts, choose business positioning without an owner, publish metadata, or decide that a disputed claim is true.

When to Use

• A person, organization, brand, method, product, service, or offer is confused with another.
• Names, descriptions, roles, dates, attributes, or relationships vary across surfaces.
• Search or generative systems need stable target entities and canonical claims.
• Structured or machine-readable representation is being considered.
• Material authority, outcome, credential, or differentiator claims need provenance.

When NOT to Use

• The task is only terminology cleanup inside one codebase; use domain-language-alignment.
• The primary question is whether a knowledge source may enter governed context; use knowledge-source-evaluation.
• The entities and claims are already stable and the dominant need is crawl/index review, generative measurement, or action execution.
• The requester expects the skill to settle a legal, ownership, brand, or policy dispute.

Core Moves

1. Bound the representation decision. Name the target property, audiences, entity types, affected surfaces, decision owner, required freshness, and evidence currently available.
2. Model entities and relationships. Identify canonical names, aliases, types, identifiers, roles, attributes, versions, and relationships that must remain distinguishable.
3. Compare representations. Inspect available pages, metadata, structured data, profiles, repositories, and approved external references without treating absence as proof.
4. Connect claims to evidence. Separate facts, inferences, hypotheses, recommendations, conflicts, and unavailable evidence; record source, method, date, owner, and limitations.
5. Produce a bounded representation handoff. Recommend canonical descriptions, relationship corrections, ownership, verification, and downstream work without mutating consumer surfaces.

Optional Modules

• Entity inventory — Inventory entities, aliases, types, identifiers, relationships, and authoritative surfaces when several entities or historical names exist.
• Canonical description design — Draft short, medium, and detailed source-backed descriptions when systems or readers receive inconsistent explanations.
• Entity consistency audit — Compare names, dates, roles, categories, attributes, and relationships across multiple surfaces.
• Claim-evidence mapping — Connect material claims to evidence, dates, owners, and limitations when authority, expertise, outcomes, credentials, numbers, or differentiation matter.
• Knowledge-graph readiness — Check whether entity types and relationships are explicit enough for graph-like interpretation when a concrete representation consumer exists.
• Machine-readable packaging — Evaluate schema, feeds, sitemaps, documentation forms, or other packaging only when a named consumer or retrieval problem justifies it.
• Version and freshness governance — Distinguish current, historical, deprecated, and time-sensitive representations when offerings or organizations evolve.

Activation Triggers

• Activate entity inventory when more than one person, brand, method, organization, offering, alias, or version appears.
• Activate canonical description design when descriptions vary materially or blur authority, brand, method, product, or offering.
• Activate entity consistency audit when representations span several owned or external surfaces.
• Activate claim-evidence mapping when a material claim influences trust, eligibility, comparison, or a consequential decision.
• Activate knowledge-graph readiness only when structured entity reconciliation is a stated objective.
• Activate machine-readable packaging only when a concrete consumer and failure mode are named; no tactic is mandatory by default.
• Activate version and freshness governance when descriptions, offers, teams, facts, or evidence change over time.

Expected Output

entity_representation_review:
  scope:
    target_property: <property or corpus>
    decision: <representation decision>
    owner: <human owner or unknown>
    observed_at: <date or unavailable>
  entities:
    - id: <stable local id>
      canonical_name: <name or unresolved>
      type: <person | organization | brand | method | product | service | other>
      aliases: [<alias>]
      relationships: [<typed relationship>]
      authoritative_surfaces: [<source>]
      freshness: <current | historical | deprecated | unknown>
  findings:
    - finding_id: <stable id>
      skill: knowledge-entity-representation
      module: <activated module>
      target: <entity, claim, relationship, or surface>
      finding_type: blocker | degradation | risk | opportunity | unknown
      evidence:
        source: <source or unavailable>
        observed_at: <date or unavailable>
        method: <inspection method>
        raw_observation: <observation or unavailable>
      confidence: high | medium | low | unavailable
      fact: <source-backed fact or unavailable>
      inference: <bounded interpretation or none>
      hypothesis: <testable proposition or none>
      impact: <affected understanding or decision>
      recommendation: <bounded recommendation>
      verification: <observable check>
      handoff: <owner or skill>
  canonical_guidance:
    descriptions: [<source-backed description candidate>]
    relationship_corrections: [<correction>]
    unresolved_conflicts: [<conflict or missing evidence>]
  limitations: [<scope or evidence limit>]

Verification

• Every entity has a stable local identifier or is explicitly unresolved.
• Canonical names, aliases, types, roles, relationships, versions, and authoritative surfaces are source-backed or marked unknown.
• Facts, inferences, hypotheses, recommendations, conflicts, and unavailable evidence are distinct.
• Material claims identify evidence, observation date, method, owner, and limitations.
• Proposed descriptions preserve human clarity and do not hide disagreement behind polished copy.
• Machine-readable packaging is recommended only for a concrete consumer and supported type.
• The output names verification and ownership without making external changes or business decisions.

Handoff Signals

• The representation problem is primarily wording or terminology inside one domain → domain-language-alignment or documentation.
• Source eligibility or precedence is unresolved → knowledge-source-evaluation or knowledge-conflict-resolution.
• The target entities are stable and the next need is discovery/crawl review → search-indexability-optimization.
• Repeated generative representation must be measured → ai-discovery-measurement.
• A legal, ownership, credential, policy, or brand decision lacks an authorized owner → human or governance review.
• A schema, feed, page, profile, or repository must be changed → implementation owner.

Pairs Well With

• domain-language-alignment
• documentation
• knowledge-source-evaluation
• knowledge-conflict-resolution
• restricted-context-check
• observability-review
• search-indexability-optimization
• ai-discovery-measurement
• generative-visibility-optimization

Anti-patterns

• Treating a search result, generated answer, or repeated phrase as authoritative by itself.
• Inventing identifiers, relationships, credentials, outcomes, dates, or corroboration.
• Collapsing a person, organization, brand, method, product, and offering into one entity for convenience.
• Treating missing evidence as evidence of absence or falsehood.
• Resolving disputed ownership, positioning, or policy without an authorized decision owner.
• Recommending schema, llms.txt, feeds, or a knowledge graph without a concrete consumer problem.
• Converting a provider-specific observation into universal representation truth.
• Optimizing machine-readable descriptions at the expense of accurate, accessible human meaning.