| name | ralph-loop-governance |
| description | Keep Ralph operational files internally consistent with single-source decision logging. Use when updating STATE, decisions, learnings, and task trackers. |
Ralph Loop Governance
Use this skill when updating any files under .ralph/.
Core Rule
Decisions have exactly one source of truth:
- Write all D-XXX entries only in
.ralph/agent/decisions.md
- Never duplicate decision tables or detailed decision entries in
.ralph/STATE.md
- Keep exactly one active plan in
.ralph/STATE.md; mark all others as baseline/reference
File Responsibilities
.ralph/STATE.md: handoff status only (phase, progress, blockers, next steps, pointers)
.ralph/agent/decisions.md: all formal decisions with rationale
.ralph/agent/learnings.md: LRN-XXX learnings and actionable prevention notes
.ralph/tasks/*.md: task checklists and execution progress
Consistency Checklist
Before commit:
STATE.md references decisions.md but does not repeat decision content
- Any new decision in session appears in
decisions.md
- Any process mistake/pattern appears in
learnings.md
- Task status files reflect active vs baseline runs correctly
- No contradictions across STATE, tasks, decisions, and learnings