platform bets

Platform bets

These are separate products, not commitments on the current roadmap. None starts until Phases 1–3 of the roadmap are in daily use. Each item needs its own scoped design before implementation.

Item	Why it is late	Honest constraint
Worker inside bun build --compile	Binaries already degrade to in-process	Needs a Bun embedding incantation that actually resolves at runtime; fallback stays
Structured-clone benchmark / coalesce stats echo	Measure, not defect	Source/npm only
Remote / SSH / Dev Container supervisor	Breaks “loopback + one socket per checkout”	New process model, auth on RPC
Multi-repo TUI	Two checkouts are two daemons by design	Picker that attaches to another repoID
K8s / Skaffold import	Cluster objects have no local equivalent	Keep discovery hints only unless we invent a tiny subset
Container build, networks, --workdir	Compose parity	Schema + containers.ts only; no k8s. Memory/CPU/PIDs defaults and optional user / read_only / cap_drop already exist
OIDC browser / device / refresh persistence	Plugin is client-credentials by design	New plugin, not core Google
OIDC as route auth	Proxy auth is none / iap / service_account	New proxy adapter path
Non-Google SSO in core	Violates “Google is an adapter” if it lands in domain	Plugin only
Windows named-pipe DACL	Bun does not expose DACL	Pipe ACL still undocumented; RPC frames require rpc-token from the user profile so other Windows users cannot drive the supervisor
Apple notarization / Authenticode	Release/legal, not app code	Signing pipeline in .github
Custom TUI layouts, crash bell, mouse drag-select	Chrome, not orchestration	After parity
Workflow DAG beyond tasks + start waves	Tasks already exist	Only if /run pickers prove insufficient

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

Related

• Roadmap
• Architecture
• Plugins
• MCP