ralph-multimodel

---

LIBRARY-FIRST PROTOCOL (MANDATORY)

Before writing ANY code, you MUST check:

Step 1: Library Catalog

• Location: .claude/library/catalog.json
• If match >70%: REUSE or ADAPT

Step 2: Patterns Guide

• Location: .claude/docs/inventories/LIBRARY-PATTERNS-GUIDE.md
• If pattern exists: FOLLOW documented approach

Step 3: Existing Projects

• Location: D:\Projects\*
• If found: EXTRACT and adapt

Decision Matrix

Match	Action
Library >90%	REUSE directly
Library 70-90%	ADAPT minimally
Pattern exists	FOLLOW pattern
In project	EXTRACT
No match	BUILD (add to library after)

---

STANDARD OPERATING PROCEDURE

Purpose

Run multi-model RALPH flows that leverage specialized agents for reasoning, alignment, learning, planning, and handoff with controlled synthesis.

Trigger Conditions

• Positive: problems needing diverse model strengths, cross-model validation, parallel evidence gathering, and adjudicated synthesis.
• Negative: single-model work, prompt-only edits (route to prompt-architect), or new skill creation (route to skill-forge).

Guardrails

• Skill-Forge structure-first: keep SKILL.md, examples/, tests/ current; add resources//references/ or note gaps.
• Prompt-Architect hygiene: capture HARD/SOFT/INFERRED constraints per model/phase, avoid VCL leakage, and publish ceilings for confidence.
• Multi-model safety: assign roles, enforce registry usage, prevent uncontrolled self-calls, and keep hook latency within budget.
• Adversarial validation: run cross-model disagreement checks, COV per synthesis, and boundary tests; document evidence.
• MCP tagging: store runs with WHO=ralph-multimodel-{session} and WHY=skill-execution.

Execution Playbook

1. Intent & roster: define objective, select models/roles, and confirm constraints.
2. Phase wiring: map RALPH phases to models, timeboxes, and success metrics.
3. Deliberation: gather model outputs, run challenges, and update shared evidence.
4. Synthesis: reconcile disagreements, choose outputs, and plan handoff with rollback paths.
5. Validation loop: stress-test synthesis, measure performance, and log telemetry.
6. Delivery: share decisions, evidence, risks, and confidence ceiling.

Output Format

• Objective, constraints, and model roster with roles.
• Phase summaries, evidence, and dissent.
• Handoff/rollback plan and risk register.
• Confidence: X.XX (ceiling: TYPE Y.YY) - rationale.

Validation Checklist

• Structure-first assets present or ticketed; examples/tests reflect multi-model paths.
• Role boundaries enforced; registry-only agents used; hook budgets verified; rollback ready.
• Adversarial/COV runs logged with MCP tags; confidence ceiling stated; English-only output.

Completion Definition

Flow is complete when synthesis is chosen with evidence, handoff executes, risks are owned, and logs persist in MCP with session tags.

Confidence: 0.70 (ceiling: inference 0.70) - Multi-model RALPH doc aligned to skill-forge scaffolding and prompt-architect evidence/confidence discipline.