🤖 curated
From Discord: "TODO: let's move the npm package to https://npmjs.com/package/framework".
The name is already ours. framework on npm lists brillout and suleimansh as its maintainers. It was created by Mikeal Rogers in 2011 and last really published as 0.6.0 in April 2022; the registry record was modified 2026-08-11, which is the handover. So this is a rename, not an acquisition — no dispute process, no waiting on npm support.
Where we are
|
|
| Published today as |
@gemstack/the-framework |
| Latest on npm |
1.5.0, published 2026-08-19 (7 versions) |
package.json says |
"version": "0.0.0" |
Legacy versions under framework |
up to 0.6.0 (2022) |
Versioning is not a problem. We are already at 1.5.0, comfortably above the 2022 0.6.0, so the first publish under framework supersedes the old package cleanly. Nothing needs deprecating or unpublishing.
Nothing in the repo records what shipped. package.json is pinned at 0.0.0 while npm serves 1.5.0, so the version is decided at publish time and the repo has no memory of it. Tolerable while a human runs the command, which is what we are keeping — see the decision below.
The work
One sub-question left open: the bin is the-framework, so the command users type stays the-framework even though the package becomes framework. npx framework would then not work, which is mildly surprising. I would keep the-framework as the command — it is what the CLI, the docs and the website all say, and renaming it is a much larger change than renaming the package. Say if you would rather they match.
Decision: publishing stays manual
It used to be automatic. .github/workflows/release.yml published on every push to main via changesets, and it was deliberately deleted in a745e2e3 — "delete the release history — changesets, changelogs, migration notes, semver" — which removed 77 changesets, the changesets tooling, four CHANGELOGs totalling 2,861 lines, and reset the versions to 0.0.0.
We are leaving it deleted. The reasons, in order of weight:
- Nothing installs the package. No workflow does, and a dogfood runs
node packages/the-framework/dist/bin.js from the local build. The only consumer is the README telling a human to npm i -g. A stale npm blocks no machine.
- Automation needs an
NPM_TOKEN in CI — publish rights to a freshly acquired name, sitting in a repo secret, bought for a benefit nobody is waiting on.
- It would need a version policy first.
0.0.0 in the repo versus 1.5.0 on npm has to be resolved before a workflow can bump anything. A human typing the number does not need that resolved.
- This repo merges several PRs a day. "Every green
main publishes" means several versions a day of a package with no installs.
This is a "not yet", not a "no". The day the package has users, automatic publishing is obviously right, and this issue is where that gets revisited.
The part that argues for doing this now: changesets are half-back
a745e2e3 removed the tooling but not the habit. Eleven .changeset/*.md files are tracked in the repo today, the newest added 2026-08-19 in #1595. Nothing consumes them: no .changeset/config.json, no @changesets/cli dependency, no release workflow, and no prompt instructing agents to write them. Agents write them from training, every PR commits more, and no process ever reads one.
So we currently carry the cost of changesets and none of the benefit. They go with this issue.
🤖 curated
From Discord: "TODO: let's move the npm package to https://npmjs.com/package/framework".
The name is already ours.
frameworkon npm listsbrilloutandsuleimanshas its maintainers. It was created by Mikeal Rogers in 2011 and last really published as0.6.0in April 2022; the registry record was modified 2026-08-11, which is the handover. So this is a rename, not an acquisition — no dispute process, no waiting on npm support.Where we are
@gemstack/the-frameworkpackage.jsonsays"version": "0.0.0"framework0.6.0(2022)Versioning is not a problem. We are already at 1.5.0, comfortably above the 2022
0.6.0, so the first publish underframeworksupersedes the old package cleanly. Nothing needs deprecating or unpublishing.Nothing in the repo records what shipped.
package.jsonis pinned at0.0.0while npm serves 1.5.0, so the version is decided at publish time and the repo has no memory of it. Tolerable while a human runs the command, which is what we are keeping — see the decision below.The work
name:@gemstack/the-framework→frameworkversion:0.0.0→0.7.0(brillout, 2026-08-22 — stay in 0.x for experimental semver)homepage,bugs,repository: all three still point atgemstack-land/the-framework, and follow the repo toframework/the-frameworkREADME.md— the npm badge and both install lines (npm i -g @gemstack/the-framework,npx @gemstack/the-framework).changeset/*.mdfiles (see below)One sub-question left open: the
binisthe-framework, so the command users type staysthe-frameworkeven though the package becomesframework.npx frameworkwould then not work, which is mildly surprising. I would keepthe-frameworkas the command — it is what the CLI, the docs and the website all say, and renaming it is a much larger change than renaming the package. Say if you would rather they match.Decision: publishing stays manual
It used to be automatic.
.github/workflows/release.ymlpublished on every push tomainvia changesets, and it was deliberately deleted ina745e2e3— "delete the release history — changesets, changelogs, migration notes, semver" — which removed 77 changesets, the changesets tooling, four CHANGELOGs totalling 2,861 lines, and reset the versions to0.0.0.We are leaving it deleted. The reasons, in order of weight:
node packages/the-framework/dist/bin.jsfrom the local build. The only consumer is the README telling a human tonpm i -g. A stale npm blocks no machine.NPM_TOKENin CI — publish rights to a freshly acquired name, sitting in a repo secret, bought for a benefit nobody is waiting on.0.0.0in the repo versus1.5.0on npm has to be resolved before a workflow can bump anything. A human typing the number does not need that resolved.mainpublishes" means several versions a day of a package with no installs.This is a "not yet", not a "no". The day the package has users, automatic publishing is obviously right, and this issue is where that gets revisited.
The part that argues for doing this now: changesets are half-back
a745e2e3removed the tooling but not the habit. Eleven.changeset/*.mdfiles are tracked in the repo today, the newest added 2026-08-19 in #1595. Nothing consumes them: no.changeset/config.json, no@changesets/clidependency, no release workflow, and no prompt instructing agents to write them. Agents write them from training, every PR commits more, and no process ever reads one.So we currently carry the cost of changesets and none of the benefit. They go with this issue.