Skip to content

Platform bets (Phase 5)

These are separate products. Do not start them until Phases 1–3 of the product-gaps roadmap are in daily use. Each item needs its own scoped design.

ItemWhy it is lateHonest constraint
Worker inside bun build --compileBinaries already degrade to in-processNeeds a Bun embedding incantation that actually resolves at runtime; fallback stays
Structured-clone benchmark / coalesce stats echoMeasure, not defectSource/npm only
Remote / SSH / Dev Container supervisorBreaks “loopback + one socket per checkout”New process model, auth on RPC
Multi-repo TUITwo checkouts are two daemons by designPicker that attaches to another repoID
K8s / Skaffold importCluster objects have no local equivalentKeep discovery hints only unless we invent a tiny subset
Container build, networks, limits, --workdirCompose paritySchema + containers.ts only; no k8s
OIDC browser / device / refresh persistencePlugin is client-credentials by designNew plugin, not core Google
OIDC as route authProxy auth is none / iap / service_accountNew proxy adapter path
Non-Google SSO in coreViolates “Google is an adapter” if it lands in domainPlugin only
Windows named-pipe DACLBun does not expose DACLDocument until Bun can; optional RPC token is a bigger threat-model change (Unix also trusts anyone who can open the socket)
Apple notarization / AuthenticodeRelease/legal, not app codeSigning pipeline in .github
Custom TUI layouts, crash bell, mouse drag-selectChrome, not orchestrationAfter parity
Workflow DAG beyond tasks + start wavesTasks already existOnly if /run pickers prove insufficient

Interactive gcloud login stays a local TTY flow. MCP never owns a browser or a login TTY.

Released under the MIT License.