refactor(storage): drop the barrel and publish narrow entrypoints

[childrentime]

Summary

maka --help printed Node's SQLite ExperimentalWarning to stderr, on a command that never opens a database. Every other CLI entry did the same. The cause is structural rather than local, so this removes the structure instead of muffling the symptom: @maka/storage no longer publishes a barrel.

Fixes the regression first reported in apache#1257.

Why the warning was there

Node evaluates a builtin the moment it enters a module graph. A static value import is therefore not a declaration of intent — it is the load:

import { DatabaseSync } from 'node:sqlite'; // evaluated here

Exactly three modules do this: operational-target-schema.ts, operational-state-backup.ts and session-bundle-policy.ts.

operational-target-schema.ts is the amplifier. operational-state-store.ts imports it — and operational-state-store.ts is imported by roughly forty modules, which is how three import statements reached 45 of the package's 110 modules and 24 of the barrel's 43 export entries. operational-state-store.ts already loaded SQLite lazily; it was pulled under anyway by the module it imports.

The last link is the CLI. cli-core.ts imports @maka/storage to reach resolveMakaDataRoots, a pure path helper. The barrel's export * surface hands it the whole storage layer, SQLite included, before --help prints a single line.

Change

• Delete packages/storage/src/index.ts (159 lines) and drop . from the exports map.
• Publish the 21 modules consumers actually reached through the barrel as narrow subpaths, for 52 entrypoints total.
• Rewrite 87 files onto those subpaths. The 20 touched files in packages/runtime are all tests; that package's production code is untouched.
• Two tests moved off the barrel's shape rather than its contents. managed-workspace-baseline asserted that internals were absent from the barrel object and now asserts that their modules are absent from the exports map — a stronger claim, since reachability is now decided there. provider-request-capture-artifact reaches its subject directly.

Regression guard

public-entrypoints.test.ts imports all 52 published entrypoints concurrently (0.6s) and asserts:

• the package publishes no . entrypoint;
• the entrypoints that load node:sqlite are exactly the 29 listed in the test, leaving 23 that are free of it.

That list is the package's SQLite boundary written down. Widening it later means editing the list and saying why.

This matters because of how apache#1257 came back. Its guard, package-import.test.ts, asserted an empty stderr on the package entrypoint; apache#2710 removed it in a bulk test cleanup, apache#1994 then added two of the static imports and apache#2445 a third, and the warning shipped. A list of entrypoints is harder to delete without noticing than an assertion that something is empty.

Validation

Node v24.11.1, macOS arm64. Before/after measured on the same source tree by checking main out, building, running, and switching back.

Command	main	This branch
maka --version	169 B stderr	0 B
maka --help	169 B stderr	0 B
maka run --help	169 B stderr	0 B
maka run <prompt>, isolated profile	169 B stderr	0 B

The last row is the case laziness alone cannot cover, because a real session does open a database. It stays clean because the Runtime Host that owns the database is a child process spawned with stdio: 'ignore' (packages/runtime-host/src/client/launcher.ts), and packages/cli/src opens no database itself. In the isolated profile used for that row, the host created and initialised a 917 KB runtime.sqlite while the client's stderr stayed empty.

Check	Result
npm run build	pass
npm run typecheck	pass
npm run lint	pass, 2452 files
npm test --workspace maka-agent	315 tests, 315 pass
npm test --workspace @maka/runtime-host	1019 tests, 1019 pass
npm test --workspace @maka/storage	853 tests, 838 pass, 14 skipped, 1 pre-existing failure (identical on main)

Remaining warning sites

The paths that still emit the warning all genuinely use SQLite, so the warning is expected there:

• maka eval (snapshots result databases) and maka activate (validates session bundles against SQLite files) — including their --help, since the command module loads before argument handling;
• the 29 SQLite-backed entrypoints of @maka/storage;
• the processes that actually open databases (the Runtime Host child, the Electron main process).

Compatibility

Dropping . is a breaking change for anything importing the bare specifier. @maka/storage is not published independently and every consumer lives in this repository, so the blast radius today is zero — but if the package is ever published on its own, this belongs in the release notes.

Generative tooling disclosure

Claude Code (Claude Fable 5) made a substantive contribution: it traced the regression to the barrel, performed the mechanical rewrite, wrote public-entrypoints.test.ts, ran the before/after measurements, and drafted this description. Commits carry the Generated-by: Claude Code trailer.

A human contributor of record has reviewed the diff, verified the evidence above and decided to submit it.

[childrentime]

Submitted upstream as apache#3301. Closing this review copy.