Add deliverability judge skill

[pay4love1]

Adds a read-only deliverability-judge runx skill package for Frantic bounty #65.

Included files:

• skills/deliverability-judge/SKILL.md
• skills/deliverability-judge/X.yaml
• skills/deliverability-judge/run.mjs
• skills/deliverability-judge/fixtures/*
• skills/deliverability-judge/harness-evidence/local-output.md

Behavior:

• Fuses sealed postmaster, bounce, complaint, and placement signals against operator policy thresholds.
• Emits a read-only continue, throttle, or pause recommendation only when signals are complete, sealed, and non-contradictory.
• Refuses to emit a recommendation and escalates when signals are missing, unsealed, malformed, or contradictory.
• Does not mint authority, hold state, emit effects, or trigger a send/throttle action.

Local direct-run evidence is included in harness-evidence/local-output.md. The inline harness cases are sealed_healthy_signals_continue and contradictory_signals_escalate.

[pay4love1]

Status update for Frantic bounty #65 (deliverability-judge).

Implemented and pushed the skill package, fixtures, local direct-run evidence, evidence.json, verification.json, and report.md.

Current blocker is the runx publish auth flow rather than the skill logic:

• runx-cli 0.6.14 is available locally.
• runx login --provider github --for publish generates connect.runx.ai sessions.
• The operator manually retried the GitHub sign-in flow, but after GitHub password sign-in the connect page returns 404 Not Found.
• Because publish auth cannot complete, the package is not yet in the runx registry and no sealed dogfood receipt_ref exists yet.

The PR artifacts are public and pinned in the latest head commit. I can complete registry publish, dogfood, verify, and Frantic delivery as soon as the runx Connect callback/auth issue is cleared.