Health · Risk radar

健康度メーター詳細。5 指標 + 60/90/120 分ガードレール連動。 spec: specs/portal/specs/health-meter.md §5 + §60/90/120 連動

Overall
WARN worst-of
Non-green
1 indicators
Owner elapsed
15m fresh
Spec ref
✓ 準拠

Health · Risk radar

WARN
  • Hook latency p50 good
    87 ms <200ms / 200-1000ms / >1000ms

    median over last 1h, target <200ms

  • Sub-agent parallel good
    2 AIs ≤3 / 4-5 / >5

    last 5min: claude main + codex reviewer

  • Session length proxy warn
    72000 tok <50k / 50-100k / >100k tok

    estimated, events × 200; 5-10万で黄、handoff 候補

  • MCP failure rate (24h) good
    2.3 % <5% / 5-20% / >20%

    4 errors / 174 calls last 24h; hubspot-dev zombie

  • Mistakes inbox backlog good
    3 pending <10 / 10-30 / >30

    01_受信箱; 0 high severity, 0 over-30d

Spec: 99_system/architecture/v4/specs/health-meter.md §5 指標 / overall = max(severity)

60 / 90 / 120 分 Guardrail

fresh
  • 60m checkpoint

    progress.md 更新提案

  • 90m soft stop

    新規 sub-agent 禁止、統合に寄せる

  • 120m hard stop

    handoff 更新義務 + 続行可否を AI に問う

現状: 15m elapsed

Trend (24h)

placeholder
hook_latency_p50
87 ms
Sparkline (Stage E): ▁▁▂▁▁
subagent_parallel
2 AIs
Sparkline (Stage E): ▁▁▂▁▁
session_length_proxy
72000 tok
Sparkline (Stage E): ▁▁▂▁▁
mcp_failure_rate
2.3 %
Sparkline (Stage E): ▁▁▂▁▁
mistakes_inbox_backlog
3 pending
Sparkline (Stage E): ▁▁▂▁▁

Stage E で hooks.log の直近 24h を時系列 sparkline 化予定。spec: specs/health-meter.md §クリックで詳細パネル