Lytics Data Health

Answers "is my data flowing correctly?" Two read-only modes: a unified health report across streams, jobs, schema, and event quota, and a stream inspector for listing streams, their stats, and recent events. Nothing here writes to the account.

Before you start

• Credentials: references/auth.md (LYTICS_API_TOKEN, optional LYTICS_API_URL).
• Calling conventions, the /v2 vs /api response shapes, and status-code handling: references/api.md.
• Every endpoint in this skill is a GET; run them without asking. If the user then wants a fix (bounce a job, mark an identifier), hand off to the owning skill, whose writes go through references/confirmation-gate.md.

Gotchas

• An old job updated is not a stuck job. updated is when the job's config was last edited; every healthy long-running job looks stale by it. Judge progress by the timestamp of the job's most recent log event (GET /v2/job/{id}/logs). Never recommend bouncing a job on staleness alone.
• Job status values are running, initializing, sleeping, paused, pausing, fault-N (prefix fault: erroring or backing off after N errors), failed, completed, deleted, deleting. A v2 kill is a delete, so killed jobs show as deleted/deleting.
• GET /v2/job hides only completed and killed (deleted) jobs by default -- failed and fault-N jobs are already in it. Add ?show_completed=true for completed ones, ?show_all=true for killed ones.
• Field freshness is per-field last_seen on GET /api/schema/_streams, not the schema field's modified, which is when its definition was last edited.
• One failed check must not block the report: report the error for that dimension and continue with the others.

Modes