[cpitzi]

Summary

Regenerates fleet-reports/epics.md from live state after #287 merged and its Terraform apply created the epic label on every repo. The 23 tracking issues the first render could not see (the checklist-only campaigns such as #276, #237 and #257, the four 2026-07-04 roadmap issues, and the epics closed since August) now carry the label, so the report shows 14 open and 9 closed epics instead of 10 and 3, and attributes 231 of the month's 620 merged PRs to an epic (44% of non-bot work, up from 22%).

Report-only change: the generator, workflow and docs are unchanged from #287. The Monday workflow would have produced the same refresh; this lands it today so the report reflects the labelled state.

🤖 Generated with Claude Code

[github-actions]

Review (haiku, claude-review.yml)

Review of PR #288: Epics report refresh (fleet-reports/epics.md)

The diff touches only the generated fleet-reports/epics.md. Terraform, CI, fleet-ops/, brand and the community-health files are untouched, so the live-apply, fleet-config and brand checks have nothing to act on. The internal arithmetic holds: 14 open and 9 closed epic rows, 48 open issues outside epics, and 231 of 521 non-bot PRs (44%). I could not confirm the live epic label count, because gh search was blocked in this session.

1. minor, PR body: No Closes #<n>, Fixes #<n> or Resolves #<n> line, so fleet PR-workflow item 1 is not met. The body also names no tracking issue for the refresh, so either an issue should be linked or the omission should be stated.

2. minor, fleet-reports/epics.md:295-299: The new rename-debt and cluster entries put on-prem detail into a public file: the cluster="lentago-lab" label value, homelab-cluster, n8n.lab.lentago.dev, and LXC numbers (e.g. lines 292, 299). The CLAUDE.md policy exception for on-prem architecture covers only incident reports. The sibling fleet report is "public metadata, no on-prem-internal detail", and the epics report is meant to be its public sibling. The generator copies issue titles verbatim, so the fix belongs in the generator or the source titles rather than in this file. The scope call needs confirming before the next weekly refresh republishes it.

3. minor, fleet-reports/epics.md:342-344: The Fleet secrets epic publishes that FLEET_REPORTS_TOKEN is over-broad, is due to expire on 2026-10-30, and needs re-minting. That is a credential's scope and expiry in a public file. No token value appears, so there is no leak, but the report advertises the weakest credential the weekly job depends on. Confirm that this level of detail is intended for a public report.

[cpitzi]

Review triage (all three findings):

1. Declined. A report refresh has no issue to close, the same as the weekly fleet-reports-refresh PRs the workflow opens. Stated here so the omission is on record.
2. Declined, scope confirmed. The report publishes issue titles verbatim, exactly as the sibling fleet-report.md already does for every open issue (the same titles appear there today: the homelab-cluster rename, the cluster="lentago-lab" migration, the LXC 105 producer). Every title is public on its issue. The hard line in CLAUDE.md is a credential, a key, a private IP-plus-purpose map, or a secret; none appears. No generator change.
3. Declined, same ground. FLEET_REPORTS_TOKEN's name, scope question and expiry are the titles of Re-mint two over-broad GitHub tokens: FLEET_REPORTS_TOKEN and the workstation PAT #242 and [renewal] FLEET_REPORTS_TOKEN — fine-grained PAT expiry — due 2026-10-30 #271, both public issues already listed in fleet-report.md, and [renewal] FLEET_REPORTS_TOKEN — fine-grained PAT expiry — due 2026-10-30 #271 is opened on purpose by the public renewal calendar (renewal-calendar.yml). As Credential register: every fleet credential's store, scope, issuer and expiry (names only) #243 records, secret names and their stores are already public in the workflow files; no value appears anywhere. Flagged to the maintainer for awareness rather than changed here.

Count check: 48 open issues outside any epic, 80 under one, 14 open epics; 80 + 48 + 14 = 142 open issues in the public fleet. The reviewer's figures match the file.