AIF — End-to-End Methodology Flow

Run New_Framework_Setup_Agent.md — creates folder structure, copies CLAUDE.md template, scaffolds all required subfolders, creates ProjectStatus.md, appends ENGINE_CHANGELOG entry. Takes approximately 10 minutes.

Run six pre-work agents in sequence to produce the seven workshop deliverables. Each agent output is reviewed by the facilitator before use. The Agent Primer is the only item sent to participants before the workshop (2–3 days prior).

Complete the 11-item readiness checklist from L1_design_framework.md Section 6 before the main workshop sessions can proceed. All pre-work deliverables must be complete and facilitator-reviewed, and a settling-in session (Phase 2.5, sponsors/leads only) must have been held — new as of 2026/08, to give sponsors/leads time to get into the right decision-making head space ahead of the wider workshop. Participants must have confirmed attendance and received the Agent Primer.

2026/08 update: runs as 3–4 sessions of 1.5–2 hours each, spread across separate days, rather than a single contact day — giving SME participants time to settle into agentic-framework thinking between sessions. Four blocks: Block 1 (condensed framing + industry review), Block 2 (foundational decisions part 1), Block 3 (HITL/autonomy decisions + agentic architecture sketch), Block 4 (open items and read-back). Drives the five foundational decisions: macro process architecture, primary orchestration unit, process classification (value stream / cross-cutting / enterprise), mandatory human oversight requirements, and near-term vs. target autonomy levels. Process Recorder captures all decisions live in one continuous Decisions Log across all four blocks.

Package workshop outputs, finalise the decisions log (all open items resolved or deferred with owner and deadline), produce the L1 Workshop Summary Deck, and confirm the L2 handoff artefact set is complete. Do not begin L2 pre-work until the decisions log is finalised.

Run L2_process_design_agent.md for each macro process to produce the sub-process architecture: 4–10 sub-processes with full SIPOC for both Interim and Target states, agent/human role assignments, HITL counts. Mandatory 3-step gate sequence before any PPTX is committed.

Review workshop: distribute draft L2 document and PPTX at least 48 hours before. Facilitate review session with domain experts and stakeholders. Capture corrections and decisions. Revise document post-session. Version increment required before use as L3 input.

Run L3_process_design_agent_V3.md for each sub-process. Produces step-level design with swimlane diagrams, agent system prompts, signal architecture, HITL register for both states. Session split protocol: if ≥5 sub-processes expected, plan separate sessions. Four-validator sequence mandatory. Six output files per sub-process.

After all L3 sub-processes for a macro process are complete, run L2_post_L3_agent_architecture_design.md to produce the eight architecture artefacts. Artefact E (interaction diagrams) uses the generate_interaction_diagrams.js script with 23-check validator. All E files use the 04_ file prefix distinguished by AGENTCODE.

After artefacts A–H are complete, run L2_post_L3_update_L2_outputs.md to reconcile L2 documents with the finalised agent architecture. Three-step process: prepare agent materials → conduct review session → apply updates. Mandatory trigger: if L3 work revealed any agent changes, SIPOC updates, or new HITL gates not in the L2 document.

Update the Framework Agent Register (FAR) to reflect the completed macro process. Update or confirm the L1 Agentic Architecture Overview once sufficient macro processes are complete. Review agent naming, tier classification, and HITL criteria for consistency across macro processes. Normalise any divergences before proceeding.

Produce three cross-framework summary deliverables. Each has a mandatory gate sequence. Run when all macro processes for the framework are complete (or at agreed milestones). Explorer has internal and client modes — never distribute internal builds externally.

Run RAISE_generation_agent.md to produce the RAISE platform configuration document. Four-step gate sequence: generate JSON → validate JSON → generate DOCX → validate DOCX. Maps all agents, sub-processes, and HITL patterns from the framework's L2/L3 designs onto the RAISE Agent Builder palette.

Produce the framework-level Control Plane Architecture document. This is a cross-process, framework-level document covering the persistent operational layer — five layers (Config, Signal, State, Intelligence, Governance) — that integrates humans, agents, process, data, and outcomes. Can begin alongside L1/L2 design; confirmed after all macro processes complete.

Cross-cutting review of the complete framework with key stakeholders. Three-step: (1) prepare a review pack covering the full L1/L2/L3 design, the Explorer, the Infographic, and the FAR; (2) conduct the review session with the client/domain sponsor; (3) resolve any final corrections and update the relevant documents. Trigger: all macro processes complete or substantially complete.

Final gate confirming the framework is complete and all deliverables are in place. Update ProjectStatus.md to mark the framework as complete. All L2 and L3 designs signed off; Explorer, Infographic, and RAISE outputs validated; HITL Master Registers complete for all macro processes; FAR and L1 Agentic Architecture Overview confirmed.