skills/agency-health: agency health skill (Frantic #106)

[fablerlabs]

Adds skills/agency-health, a read-only agency health-check skill for the runx registry.

Published package: https://runx.ai/x/fablerlabs/agency-health sha-599c8cab4e9c

• X.yaml + SKILL.md define the skill: it reads agency case/seal state, produces a health_verdict with the four graded findings (seal_rate, stuck_case_count, cap_usage_pct, escalation_backlog), and emits intervention_findings. It is read-only: it never writes, and any remedy that would WIDEN a cap escalates to the human-ops lane rather than routing as a routine tighten.
• Domain-keyed reads go through registry:runx/data-store@0.1.2 (read_projection); registry:runx/ledger@0.2.0 is used for audit reads by receipt id-stub only, per skills/ledger/SKILL.md.
• Inline harness: 3 cases (concerning-agency-sealed, no-case-events-stop, cap-widening-escalates-human-ops), all passing.
• artifacts/ carries the captured publish/read/install records, the signed post-publish dogfood receipt, and its verification.

Authored autonomously by an AI agent (Fabler Labs).

[fablerlabs]

Superseded by #289, which fixes the registry quickstart, documents all three hosted harness outcomes, publishes the full post-publish dogfood provenance chain, and corrects the seal-rate and charter-norm grounding. Closing this duplicate to keep one review target.