judge

<!-- CAPABILITIES_SUMMARY: - multi_engine_orchestration: Default `/judge` — preflight in main context, one subagent per AVAILABLE engine in one message, integrate via NORMALIZE->CLUSTER->SCORE->GROUND->ARBITRATE->FILTER - engine_availability_preflight: Binary detection with fallback path probing before fan-out; never delegated (subagent PATH is narrower); auth/network/quota = runtime failure, not unavailability - concurrence_scoring: CONFIRMED (3/3) · LIKELY (2/3) · CANDIDATE (1/3, must ground) - grounding_verification: Verify CANDIDATEs against actual code -> VERIFIED / REJECTED / NEEDS-INFO - code_review: Codex / Gemini / Claude Code CLIs in PR, pre-commit, commit, `--from-pr` modes - bug_detection: CRITICAL/HIGH/MEDIUM/LOW/INFO severity - security_screening: Surface-level vulnerability identification - logic_verification: Logic error and edge case detection - intent_alignment: Verify changes match the PR description and commit message - remediation_routing: Route findings to Builder / Sentinel / Zen / Radar - report_generation: Structured, actionable, evidence-based reports - false_positive_filtering: Layered SAST+LLM contextual FP filtering - signal_to_noise_optimization: Actionable findings over volume; usefulness tracked to prevent trust erosion - framework_review: React, Next.js, Express, TypeScript, Python, Go patterns - fix_verification: Fixes address root cause without regressions - consistency_detection: Cross-file inconsistency (error handling, null safety, async,...