Repository navigation
docs: record that this stack-breadcrumb copy is the authoritative one - #224
Conversation
The header said where the module came from and stopped there. Accurate: it did start as taskless/taskless's `@taskless/stack-breadcrumb` package. But the fact a reader needs is the other one — this copy is now authoritative, and taskless/taskless vendors it rather than maintaining its own. Without that, nothing here tells anyone that a change made in this file silently diverges a downstream copy, or that a fix made downstream is a fork the next sync discards. The cloud team hit exactly that: their CLAUDE.md credited the module to the wrong repository, and their copy's header claimed it had been "brought into this repository" from taskless/taskless — which is them, so the sentence had become self-referential when someone copied the header across. They corrected their end; this is ours. Also records why there is no drift test, since the obvious next question after "two copies exist" is "what checks they match". Nothing does, and deliberately: this repository moves ahead of the vendored copy by design, so a drift test would fail on precisely the changes that are supposed to happen. The stakes fit — a diverged breadcrumb renders a PR body slightly differently, where a diverged rule layout writes rules to the wrong place, which is why that one has a drift test and this one has a sentence. Applied to the module, its test, and the workflow, because all three carry the same origin note and a reader can arrive at any of them.
|
@claude /review |
|
Claude finished @theCodeDrift's task in 1m 30s —— View job Review completeReview mode: incremental — read 0 prior review thread(s) before reviewing.
FindingsNone. This is a comment-only change (23 added lines, 0 deletions, no logic touched) across
Nothing else in scope stood out — no inline comments to post. |
Thanks — nothing to action. Confirming the read: comment-only, no logic touched, and the three files carry the same origin note deliberately so a reader arriving at any of them finds the ownership direction. — AI Coding Agent |
Comment-only. Three files, 23 added lines, no behaviour change.
Why
The header said where the module came from and stopped there. That part is accurate — it did start as
taskless/taskless's@taskless/stack-breadcrumbpackage. The fact a reader actually needs is the other one: this copy is now authoritative, andtaskless/tasklessvendors it rather than maintaining its own.Without that, nothing here tells anyone that a change made in this file silently diverges a downstream copy, or that a fix made downstream is a fork the next sync discards.
The cloud team hit exactly that. Their CLAUDE.md credited the module to the wrong repository, and their copy's header claimed it had been "brought into this repository" from
taskless/taskless— which is them, so the sentence had become self-referential once someone copied the header across. They corrected their end; this is ours.It also records why there is no drift test
The obvious next question after "two copies exist" is "what checks they match". Nothing does, and that is deliberate — worth writing down before someone helpfully adds one.
This repository moves ahead of the vendored copy by design, so a drift test would fail on precisely the changes that are supposed to happen. The stakes fit that: a diverged breadcrumb renders a PR body slightly differently, where a diverged rule layout writes rules to the wrong place. That is why the layout table has a drift test and this has a sentence.
Worth noting the cloud team asked us to publish a
./stack-breadcrumbexport so they could diff against it, then withdrew the request on their own reasoning: that shape is for peers who must not diverge, and this is not that. No export, nothing owed either way.Scope
Applied to the module, its test, and the workflow — all three carry the same origin note, and a reader can arrive at any of them.
node --teston the module passes unchanged; house style clean.