The framework is the company's proprietary asset. It exists because the work of decoding tacit managerial judgement is not the same as the work of building an AI agent. The first is harder, slower, and more contextual. SPDF makes the first work disciplined enough that the second becomes mechanical.
The Framework
The Skill-Process Decoding Framework
The Five Stages
Elicitation
Structured reflection that surfaces the skill the manager actually uses, not the one in their job description. Five questions are answered for each skill: where is it applied, what data signals trigger it, what decision logic runs, what are the known failure modes, and what has changed in the manager's thinking in the last year. The output is conversation; the work of the next stage is to make it precise.
Structuring
The elicited material is rewritten into machine-legible form. Heuristics are stated as conditionals. Edge cases are named. Escalation triggers are identified. The manager's voice is preserved — the agent will speak in the manager's reasoning, not in a generic AI register.
Plugin Development
The structured specification is configured into an agent that runs on a real framework. Tool connections are specified: which internal systems, which external feeds, which databases the skill needs to reach. Multi-step reasoning is decomposed where the work warrants it.
Validation
The agent is tested against real decisions the manager has already made. Outputs are compared not for correctness alone but for fidelity — does the agent reason the way the manager reasons. Gaps are fed back into Stage 1. This loop runs until the manager can sign their name to the agent's outputs.
Deployment & Refinement
The agent enters the workflow with a lightweight review mechanism. The manager monitors outputs, corrects them, and feeds the corrections back into the specification. The skill stays alive. It evolves with the manager's experience and the organisation's circumstances.
The three artefacts
The framework produces three documents that together make the skill portable, auditable, and deployable.
Artefact 01
Skill-Process Specification
A structured document, typically eight to twenty pages. It contains the manager's decoded decision logic in a form that another senior practitioner could review, challenge, and sign off. It is the artefact that survives if the agent technology changes underneath it.
Artefact 02
Agent-Readiness Diagnostic Score
A judgement, expressed as a short report, on how convertible the skill is. The score is honest about where the line falls between what the agent should decide and what the human must reserve. It prevents the false confidence that derails most internal AI projects.
Artefact 03
Deployable Agent Specification
The system prompt, tool layer specification, and review interface, ready to instantiate on Claude Agents, Hermes, or an equivalent framework. The organisation's engineering team can take this and build. Framework-agnostic by design.
The four operational layers
When deployed, the agent runs in four interacting layers.
Layer 01
Skill Core
The manager's decoded heuristics and decision logic. The digitised wisdom.
Layer 02
Tool Layer
Connections to internal systems, market data, and industry repositories.
Layer 03
Orchestration
The agentic framework managing multi-step reasoning and output assembly.
Layer 04
Human Review
The interface through which the manager keeps the skill alive and current.
Continue