is-it-really-brokenlisted

Use when a health check, audit, or monitor says something is BROKEN, before repeating that to anyone. Re-runs the check from the same context the failure came from and separates real failure from unknown, since a timeout, an HTTP 000, or a permission error means the test could not answer — not that the thing is down. Prevents reporting an outage that is actually a broken probe.
TechNickAI/hermes-skills · ★ 0 · AI & Automation · score 76

Install: claude install-skill TechNickAI/hermes-skills

# Integration Health Triage ## Overview An integration audit's `BROKEN` verdict is a **test result, not a fact**. Audits probe one way; the runtime often reads another. Relay verdicts verbatim and you send people chasing phantoms — which is how a monitoring system permanently loses its reader's trust. This skill governs the gap between "the audit said X" and "here is what actually needs your attention." Measured rate on a real run (one occasion, a fleet member's weekly Integration Health Check): 7 items came back BROKEN/PARTIAL, and **3 of the 7 were audit bugs, not outages.** Assume roughly a third of any verdict list is wrong until verified. ## When to Use - A live integration audit produced a verdict file (`verdicts.json`, `REPORT.md`) or a weekly "Integration Health Check" cron delivered findings - An owner reports "my automations aren't working" against a green status list - Someone asks "what else is broken?" / "what needs my attention?" - A cron job touching third-party services failed, hung, or went quiet - An owner reports their agent keeps mentioning "still open" items, off-topic or in the middle of unrelated work — see `references/stale-agent-open-items.md` Deep-dive reference: `references/stale-agent-open-items.md` — verifying an agent's self-reported open items against live state (plugin mtime vs. gateway start time, per-interpreter Full Disk Access, the two distinct macOS iMessage faults), and how to rewrite the stale memory entry that caused the na