saddlisted

## Runtime persistence `.principled/` (in cwd) is the natural runtime emplacement for principled-related artifacts. At intake, read whatever is there if any — prior context may inform this work. When this skill produces durable artifacts, write them to `.principled/` too. Skip if absent. ## Routing Guidance - COMPETE: 'best-of-N', 'competitive generation', 'generate multiple solutions', 'quality over speed' - EXECUTE: 'spawn subagent', 'launch agent', 'delegate this', 'implement with verification', 'run in background' - JUDGE: 'multi-judge debate', 'score against rubric', 'compare candidate solutions', 'judge panel evaluation', 'consensus scoring', 'meta-judge pipeline' - EXPLORE: 'tree of thoughts', 'explore solution space', 'generate and prune', 'ideate then narrow' - For architecture design ('supervisor pattern', 'swarm', 'coordinate agents') -> use the core `subagent-orchestration` skill. ## Decision Router IF generating multiple competing solutions with meta-judge evaluation → COMPETE mode IF spawning tasks via context-isolated subagents with optional verification → EXECUTE mode IF evaluating work with meta-judge + judge pipeline or multi-judge debate → JUDGE mode IF exploring solution space with systematic pruning (ToT) → EXPLORE mode IF designing multi-agent architecture (supervisor/swarm/hierarchical patterns) -> use the core `subagent-orchestration` skill (DESIGN mode). # Mode: COMPETE Generate 3+ solutions in parallel with meta-judge evaluation and adaptive s