◎ OS PUB Apache 2.0 ← All specifications

P139 — AIEP — Cognitive Task Decomposition Protocol

Publication Date: 2026-04-12 Status: Open Source Prior Art Disclosure Licence: Apache License 2.0 Author/Organisation: Phatfella Ltd Schema: AIEP_OS_SPEC_TEMPLATE v1.0.1 — https://aiep.dev/schemas/aiep-os-spec-template/v1.0.1

Field of the Invention

[0001] The disclosure relates to task decomposition and governance systems for AI agents within computing environments.

[0002] More particularly, the disclosure concerns a cognitive task decomposition protocol for use within an Architected Instruction and Evidence Protocol (AIEP) system, enabling complex tasks to be decomposed into governed sub-tasks with explicit dependency bindings, deterministic execution ordering, and cryptographically bound completion records.

Framework Context

[0003] This invention operates within an Architected Instruction and Evidence Protocol (AIEP) environment as defined in United Kingdom patent application number GB2519711.2, filed 20 November 2025, the entire contents of which are incorporated herein by reference.

[0004] The present invention extends hierarchical planning mechanisms defined in P72, goal vector stabilisation mechanisms defined in P50, and recursive goal tree mechanisms defined in P84 while remaining independently implementable as described herein.

Background

[0005] Complex tasks submitted to AI agents contain implicit dependencies that must be made explicit before deterministic execution ordering can be established. Without governed decomposition, agents may execute sub-tasks in undisclosed orders, introduce hidden parallelism, or silently skip sub-tasks on heuristic sufficiency judgements.

[0006] There exists a need for a deterministic task decomposition mechanism that produces explicit sub-task dependency graphs, computes deterministic execution orderings from those graphs, records sub-task completion with cryptographic binding, and fails closed when dependency constraints are violated.

Summary of the Disclosure

[0007] A computer-implemented cognitive task decomposition protocol comprises: receiving a ComplexTask record; applying a version-bound DecompositionRuleset to produce a SubTaskGraph containing subtask nodes and directed dependency edges; computing a topological execution ordering of the SubTaskGraph; executing each sub-task in its deterministic order, blocking on unsatisfied dependencies; on completion of each sub-task, computing a SubTaskCompletionHash as H(subtask_id || input_evidence_hashes || output_artefact_hash || dependency_satisfaction_record); appending a SubTaskCompletionRecord; and on completion of all sub-tasks, computing a TaskCompletionHash as H(subtask_completion_hashes || decomposition_ruleset_version); returning the completed task artefact with its TaskCompletionRecord.

[0008] If a sub-task cannot be completed due to dependency violation, the task is set to non-admissible and execution is suppressed in a fail-closed manner without partial completion.

[0009] The technical effect is modification of computing system behaviour by enforcing deterministic governed task decomposition with cryptographic completion records, enabling reproducible complex task execution audits.

Claims

[0010] A computer-implemented method for cognitive task decomposition comprising: receiving a ComplexTask; applying a version-bound DecompositionRuleset to produce a SubTaskGraph; computing a topological execution ordering; executing sub-tasks in deterministic order; computing SubTaskCompletionHashes; and returning a TaskCompletionRecord binding all sub-task records.

[0011] The method of claim 1 wherein dependency violations cause fail-closed non-admissible outcomes without partial task completion.

[0012] The method of claim 1 wherein execution ordering is computed deterministically from the SubTaskGraph without heuristic scheduling.

[0013] A system for cognitive task decomposition comprising one or more processors and a non-transitory computer-readable medium storing instructions to execute the method of any preceding claim.