Fleet signals — from GitHub to the alarm panel

Fleet signals — from GitHub to the alarm panel A data-flow diagram generated by Archify. 01 / Fleet 02 / Gather 03 / Store 04 / Derive 05 / Surface Fleet repos · issues · PRs · CI · 01 / Fleet · ~40 Fleet repos issues · PRs · CI ~40 Actions API · run history · 01 / Fleet · PyGithub Actions API run history PyGithub dash-gen triage · open-state snapshot · 02 / Gather dash-gen triage open-state snapshot dash-gen actions · minutes · waste · 02 / Gather dash-gen actions minutes · waste _data/*.yml · committed, diffable · 03 / Store · memory _data/*.yml committed, diffable memory dash-gen remediate · failing + expensive · 04 / Derive · capped dash-gen remediate failing + expensive capped dash-gen harness · scorecard + trip wires · 04 / Derive · offline dash-gen harness scorecard + trip wires offline Doctor agent · draft PR per candidate · 05 / Surface · Opus Doctor agent draft PR per candidate Opus Dash site · /triage · /harness · 05 / Surface · Jekyll Dash site /triage · /harness Jekyll open state issues, PRs, red runs run history minutes per workflow fleet_triage.yml daily snapshot actions_usage.yml daily snapshot red + expensive two signals, one queue freshness + rates all four ledgers fix queue ranked, deduped harness_health.yml scorecard + wires Legend primary data policy / PII data store data flow

Gather

  • • Generators are deterministic Python, not an LLM
  • • Every snapshot carries the generated_at the stale-data wire watches
  • • One pulse job, one checkout, one commit

Derive

  • • remediate keys on owner/repo:workflow — one candidate per workflow
  • • harness reads only committed files; reproducible anywhere
  • • Thresholds and caps are versioned in fleet.yml

Surface

  • • The doctor acts on the queue, bounded by caps
  • • The dash renders the same YAML for people
  • • Both can be replayed from git history