ralph-loop

STANDARD OPERATING PROCEDURE

Purpose

Coordinate the RALPH sequence (Reason, Align, Learn, Plan, Handoff) with explicit constraints, validation, and confidence ceilings.

Trigger Conditions

• Positive: structured iterative problem solving, alignment checkpoints, learning/reflection loops, plan-to-action handoffs.
• Negative: ad-hoc single-pass answers, prompt-only edits (route to prompt-architect), or new skill weaving (route to skill-forge).

Guardrails

• Skill-Forge structure-first: ensure SKILL.md, examples/, tests/ exist; add resources//references/ or log remediation.
• Prompt-Architect hygiene: extract HARD/SOFT/INFERRED constraints at each phase, maintain pure English, and state confidence with ceilings.
• Loop safety: enforce phase boundaries, timeboxes, and evidence requirements; use registry agents only; keep hooks within latency budgets.
• Adversarial validation: run COV after each iteration, probe wrong-turn scenarios, and capture evidence.
• MCP tagging: save loop traces with WHO=ralph-loop-{session} and WHY=skill-execution.

Execution Playbook

1. Reason: clarify the problem, constraints, and gaps; confirm inferred assumptions.
2. Align: align stakeholders/agents on goals, guardrails, and success metrics.
3. Learn: gather evidence, test hypotheses, and record findings.
4. Plan: map actions with owners, timelines, and rollback checkpoints.
5. Handoff: delegate tasks, verify readiness, and monitor execution.
6. Validation: run adversarial checks, measure outcomes vs metrics, and log telemetry.
7. Delivery: present outcomes, evidence, residual risks, and confidence ceiling.

Output Format

• Phase-by-phase summary (R, A, L, P, H) with constraints and decisions.
• Evidence log, risks, and open questions.
• Handoff plan with owners and checkpoints.
• Confidence: X.XX (ceiling: TYPE Y.YY) - rationale.

Validation Checklist

• Structure-first assets present or ticketed; examples/tests reflect RALPH flow.
• Phase gates respected; registry and hook budgets validated; rollback paths ready.
• Adversarial/COV runs logged with MCP tags; confidence ceiling declared; English-only output.

Completion Definition

Loop is complete when handoff executes against the plan, evidence confirms outcomes, risks are owned, and logs persist with MCP tags.

Confidence: 0.70 (ceiling: inference 0.70) - RALPH loop documentation aligned to skill-forge scaffolding and prompt-architect clarity with explicit evidence and ceilings.