Glossary¶
Comprehensive definitions of terms used in Task Ontology Kernel (TOK).
Core Concepts¶
Task Ontology Kernel (TOK)¶
The theoretical foundation that defines the ontological structure of tasks—identity, state, lifecycle, dependencies, and evolution. TOK transforms tasks from unstructured mental concepts into system-level primitives that can be natively executed, versioned, governed, and evolved by AI. It answers: what is a task, how does it exist, and how does it evolve.
Task-Oriented Cognitive Architecture (TOCA)¶
A cognitive architecture where tasks serve as the primary persistent unit of cognition, enabling humans and AI to collaboratively execute, evolve, and reuse structured cognitive processes. TOCA defines a five-step closed loop: Capture → Dispatch → Execute → Validate → Evolve.
Task-Native¶
A paradigm describing systems where the primary unit of execution, persistence, and evolution is a task, rather than code. In Task-native systems, code becomes a derivative artifact generated or orchestrated by tasks. This represents the fourth era of software engineering production primitives, following Machine-native, Code-native, and Service-native.
Task Object¶
The fundamental unit in TOK, composed of four layers: Intent (goal state), Context (environment and resources), Strategy (decomposition logic and tool preferences), and Evaluation (Definition of Done and validation protocol). A Task Object is machine-readable, persistent, versionable, composable, and evolvable.
Intent Layer¶
The first layer of a Task Object. Describes the goal state—what should be achieved—rather than execution steps. Must be semantically clear enough for an LLM to interpret as decision logic.
Context Layer¶
The second layer of a Task Object. Defines the environmental boundaries and resource access required for execution, including domain knowledge, historical execution records, and real-time data slots via MCP (Model Context Protocol).
Strategy Layer¶
The third layer of a Task Object. Contains the task decomposition logic and tool selection preferences. Unlike hardcoded workflows, Strategy evolves with execution experience, functioning as an adaptive "path guide."
Evaluation Layer¶
The fourth layer of a Task Object. Defines the verification protocol for task success (Definition of Done), including automated tests, Semantic Alignment checks, and human feedback loops.
Prompt Orchestration Governance (POG)¶
The paradigm that reconceptualizes prompts as persistent, governed tasks integrated into the SDLC. POG represents the conceptual shift from ephemeral prompt interactions to structured task governance.
POG Task¶
The concrete implementation of TOK in the software development domain. POG Task is the minimal executable task unit, defined in YAML, versioned via Git, and executed by AI Agents within the POG ecosystem.
POG OS¶
The operating system that implements and manages task-based cognition. It provides task runtime, memory, execution orchestration, governance, and lifecycle management.
Architecture Concepts¶
TOCA Core Loop¶
The five-step closed-loop process at the heart of TOCA: Capture (intent → task), Dispatch (task → executor), Execute (executor → result), Validate (result → evaluation), Evolve (feedback → strategy optimization).
Five-Layer Stack¶
The complete conceptual hierarchy of the POG/TOK system: Paradigm (POG) → Ontology (TOK) → Architecture (TOCA) → Primitive (POG Task) → System (POG OS).
Agent-Native Execution¶
An execution model where tasks are not directly run as scripts, but delegated to Agents who autonomously decide how to execute—selecting tools, using LLM reasoning, calling APIs, or spawning new tasks.
Execution Contract¶
The agreement formed when an Agent claims a task, defining: agent type (LLM, toolchain, hybrid), capabilities (read, write, mutate), retry strategy, and audit requirements.
Task Graph¶
A directed graph of interconnected tasks with dependency relationships (depends_on) and artifact production (produces). In the Task-native paradigm, the Task Graph replaces the traditional codebase as the primary representation of a software system.
Universal Cognitive Executor¶
A term describing LLMs' capability to execute arbitrary cognitive tasks (analysis, writing, planning, transformation) without specific code, making them the first universal task executors in history.
Intent Ontology¶
TOK's classification as an Intent Ontology—modeling "how the world will change" through executable intent—as opposed to Palantir's Object Ontology, which models "what exists in the world."