Add escalation-judge skill

[VidsHidden]

Summary

Adds the escalation-judge runx skill for Frantic bounty #69.

The skill reads a support triage packet, support thread body, prior case projection, and policy rules. It escalates only when a declared severity threshold or churn-risk signal is crossed, appends a governed case event using data-store semantics, and emits a typed escalation packet naming the downstream rail without posting or sending.

Verification

• Published package: vidshidden/escalation-judge@sha-8bdc1d12bfae
• Public URL: https://runx.ai/x/vidshidden/escalation-judge@sha-8bdc1d12bfae
• Hosted harness: https://runx.ai/x/vidshidden/escalation-judge#harness
• Hosted harness passed 3 cases:
  • sealed_high_severity_escalation
  • stop_no_threshold_no_change
  • ambiguous_low_confidence_needs_agent
• Clean install passed with runx add vidshidden/escalation-judge@sha-8bdc1d12bfae --registry https://api.runx.ai
• Direct runner evidence is committed under skills/escalation-judge/evidence/
• Windows runx skill resolves trusted registry provenance but hits the known receipt-store os error 87; .github/workflows/escalation-judge-dogfood.yml reruns the same dogfood on Ubuntu and commits durable dogfood receipt evidence.

Frantic #69 artifacts

• skills/escalation-judge/X.yaml
• skills/escalation-judge/SKILL.md
• skills/escalation-judge/run.mjs
• skills/escalation-judge/fixtures/*
• skills/escalation-judge/evidence/evidence.json
• skills/escalation-judge/evidence/verification.json
• skills/escalation-judge/evidence/report.md