BACKLOG AGENTS
— · deployed —
Auto-remediation scorecard — per-finding-kind fired / cleared / escalated / thrash over the last 7 days, mirroring backlog-agents remediation-digest. A kind that fires but never clears is detect-only (no actuator).
QUEUE (—) iBacklog items by status. In Progress counts claimed items ([~] markers + active D1 claims) — not live interactive sessions, which are their own actors under TODAY → Active Work. So Active Work can show more "live" than this counts; that's expected, not a sync bug.
WHAT NEEDS YOU VS. WHAT THE DAEMON CAN DRAIN
BLOCKERS — ITEMS WAITING ON A DECISION FROM YOU
DONE — COMPLETED ITEMS BY PROJECT iEvery [x] backlog item the fleet has finished in the last 48h, newest first, grouped by project. Older items aren't shown here — look them up with Ask (the ✦ assistant, ⌘J).
Progress against external provider budgets — Cloudflare and GitHub — so you can sanity-check spend against plan before the warning emails arrive. Snapshots refresh every ~10 min via the cron watchers.
Provider calls — outbound telemetry per cron endpoint (call counts, latest status, freshness), so you can see whether call-reduction is holding and whether a provider is failing or rate-limiting.
Work done per time bucket across the fleet — completions, tokens, or ticks, broken down by machine with a fleet-total trend line.
Historical backlog marker counts by 5h window. Reporting owns the aggregate trend; chart segments drill into WORK filters for the item rows.
Daemon vs interactive ("you") completions per day, with the daemon's share trended over time (7-day smoothed) — the read on how autonomous the fleet is becoming. iEach [x] done item is attributed by its [DONE] stamp: a precise time means an interactive session ("you"), a date-only stamp defaults to the daemon. The trend line is a 7-day rolling share, so a single day's 0%/100% spike doesn't dominate. Caveat: the timed-stamp convention is recent, so the oldest days lack interactive times and overstate the daemon share — read the recent slope, not the absolute early share.
What the fleet has been thrashing on and when it was held — aggregated from the events the daemon already records (events_store). The raw event stream is in Health → Guard.
RECENT AUDIT STREAM
Recent commit, heartbeat, and error events from fleet ticks. Reporting keeps the bounded audit scan; raw operational events stay in FLEET → Diagnostics → Guard.
LOCAL GIT HYGIENE
Uncommitted files and unpushed commits reported by fleet agents. This is a local-change hygiene snapshot, not release artifact analytics.