You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Simplify the portfolio: replace custom machinery with native options #5047
The portfolio has grown a large amount of custom machinery: roughly 300k lines of bash plus a similar amount of Go helpers and tests, and very high churn (monorepo ~1,070 commits, platform ~1,560 commits, .github 225 releases in 60 days). Much of it rebuilds something GitHub, Flux, Kyverno, Homebrew, Renovate or Dependabot already does natively. Every custom piece costs PRs, CI minutes, review rounds and agent time.
What
An audit on 2026-11-14 compared each custom implementation with its native or upstream option and kept only changes that are simpler without an unacceptable regression. Where a native option would lose a protection, the sub-issue either keeps that protection explicitly or orders the change so the stronger control lands first.
The biggest single lever is dependency-bump fan-out (one-at-a-time daily bumps, plus a .github self-pin loop that releases about once a day with no human change), followed by tests that pin prose instead of behaviour, and tool features rebuilt by hand.
The artifact publisher trust chain is handled separately (devantler-tech/.github#923) and is out of scope here.
Rejected after evaluation
Native auto-merge in place of the auto-merge workflow: may drop the untrusted-actor disarm; revisit after its settings are inventoried.
Making CodeRabbit the only review gate: loses the review-lane fallback during quota outages.
Native Claude Code worktrees and OpenTelemetry telemetry: would leave the Codex and kagent lanes uncovered.
Deploying to production after merge: loses the guarantee that main holds only deployed commits.
Replacing the workflow-shape Go tests wholesale: most of them test behaviour and the linters already run.
GUT for world-at-ruin: sound, but poor value for the effort now.
Order
Quick wins first (bump batching, signed commits, prose tests, PR-body contract, merge queue), then the medium refactors, then the strategic items.
Activity