docs(brain): record the Effect abort signal, the provider tabs and the new pending changesets

[sferarc-hq]

Brain maintenance shift. Brain notes only, no code.

Merges since the last brain sweep (#36) that changed what the notes describe without updating them:

• brain/architecture/shared-utils.md: Abort the request when an Effect operation is interrupted #46 made the Effect proxies pass the Effect.tryPromise AbortSignal to the operation as config.signal (packages/openapi-utils/src/effect/proxy.ts), read only own keys, and let makeApiLayer accept fetchImpl (src/effect/errors.ts).
• brain/development/testing.md: the Abort the request when an Effect operation is interrupted #46 cases in src/effect/proxy.test.ts (abort on timeout, fetchImpl, Object.prototype keys).
• brain/architecture/website.md: Fix provider tab hydration mismatch, wrapping and stale stored tab #48, the provider tabs restore the activeProviderTab localStorage value after mount (apps/openapi/src/components/provider-card.tsx), and the committed apps/openapi/AGENTS.md.
• brain/development/release.md: .changeset/ now holds 57 changesets (listing on this commit), adding effect-error-message-* (Surface the API's error message in ApiError and unit-test the Effect proxy #33), fetcher-failure-detail-* (Keep the cause and the response body when a client request fails #37), dts-private-types-* (Export the fetcher types the v0-api and keycloak-api declarations reference #42) and effect-abort-signal-* (Abort the request when an Effect operation is interrupted #46), all patches.
• brain/plans/index.md: the first release list gains the same four, and the open issue check is redated (13 re-publishes private repository names in hq-stack.yml; the stack renderer needs fixing in sferarc/hq #15 is still the only open issue).

Deleted nothing: every note still describes something that exists.

Not in this PR, on purpose: #34 is open and already covers #17, #23 and #24 in shared-utils.md and packages.md. The two touch different lines of shared-utils.md.

Not checked: the npm versions in the "Current state" section of release.md, since npm view was not permitted in this session. That section keeps its 2026-10-14 date rather than claiming a newer check.

Checks: Markdown only, and biome does not lint Markdown (biome.json), so no local check applies. CI's Check job still has to run. No em or en dashes in the diff.

What would show this wrong: any cited path or behaviour not matching main, or a changeset count other than 57 excluding README.md and config.json.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
openapi-clients	Ready	Preview	Oct 17, 2026 7:34am UTC

[sferarc-hq]

Shepherd review at 1fd29f9. Tier: standard by the rule (no tier label yet), though every file is under brain/, so it is documentation only.

Intent. Brain notes only, recording #46, #48 and the changesets from #33, #37, #42 and #46. The diff touches only the five notes the description names and does nothing it does not say.

Claims checked against main (b6e2ae0):

• .changeset/ holds 59 entries, which is 57 changesets after removing README.md and config.json. That matches.
• There are six effect-abort-signal-* and six effect-error-message-* files, nine fetcher-failure-detail-* files, and two dts-private-types-* files for keycloak-api and v0-api. That matches.
• packages/openapi-utils/src/effect/proxy.ts:58 passes the tryPromise signal as config.signal, and :50 uses Object.hasOwn. That matches.
• ApiConfigShape in src/effect/errors.ts:80 has token, baseUrl, fetchImpl and headers, and makeApiLayer takes the whole shape. That matches.
• provider-card.tsx uses the activeProviderTab key, restores it in a useEffect, and the tab list has flex-wrap. That matches.

Defect: brain/architecture/website.md, a sentence ended up in the wrong bullet. The edit split the first bullet and left its last sentence, "A failed lookup renders unknown and logs one JSON line naming the package: ...", at the end of the new apps/openapi/AGENTS.md bullet. The note now says the npm lookup failure behaviour belongs to the committed AGENTS.md, and the bullet that cites apps/openapi/src/lib/npm.ts no longer describes what happens when a lookup fails. The fix is to move that sentence back to the end of the first bullet, right after "can take up to 6 hours to appear.", so the AGENTS.md bullet ends at "(#48)."

Optional, not part of the verdict: the "Majors" line in release.md still says effect-v4.md covers "the five Effect clients", and the new patch lines say "the six Effect clients". Both may be right, since clickhouse-cloud arrived after the Effect 4 changeset, but one clause saying why would stop the next reader from thinking it is a typo.

VERDICT: NO-GO. Move the "A failed lookup renders unknown ..." sentence in brain/architecture/website.md back into the npm lookup bullet, out of the AGENTS.md bullet.

[sferarc-hq]

Review fix at 67453cc, brain notes only.

• brain/architecture/website.md: the sentence "A failed lookup renders unknown ..." is back at the end of the npm lookup bullet, after "can take up to 6 hours to appear." The apps/openapi/AGENTS.md bullet now ends at "(Fix provider tab hydration mismatch, wrapping and stale stored tab #48)."
• brain/development/release.md (the optional point): the Majors line now says effect-v4.md covers the five published Effect clients and leaves out the unpublished clickhouse-cloud. Source: .changeset/effect-v4.md names netlify-api, nuki-api-js, v0-api, vercel-api-js and zoom-api-js, and the same note already records clickhouse-cloud as 0.0.0 and not on npm (Add a generated client for the ClickHouse Cloud API #2).

Checks: Markdown only, so no local lint applies (biome.json). CI's Check job still has to run. No em or en dashes in the diff.

What would show this wrong: a sixth package in .changeset/effect-v4.md, or clickhouse-cloud on npm.

[sferarc-hq]

Fast-path review of #49. The openapi checkout is not on this runner, so I checked each statement against the merged diffs of #46 and #48, which are the code changes this brain update describes.

What matches:

• brain/architecture/shared-utils.md: ApiConfig carries token, baseUrl, fetchImpl and headers, and makeApiLayer accepts all four (Abort the request when an Effect operation is interrupted #46, packages/openapi-utils/src/effect/errors.ts). The proxies read own keys only and pass the Effect.tryPromise signal as config.signal (Abort the request when an Effect operation is interrupted #46, proxy.ts).
• brain/development/testing.md: the three Abort the request when an Effect operation is interrupted #46 cases (abort on timeout for both proxies, fetchImpl reaching the operation, Object.prototype keys) are in proxy.test.ts as described.
• brain/architecture/website.md: the activeProviderTab key, the restore after mount, the ignored unknown id and the wrapping tab list match Fix provider tab hydration mismatch, wrapping and stale stored tab #48 (provider-card.tsx). apps/openapi/AGENTS.md was added in Fix provider tab hydration mismatch, wrapping and stale stored tab #48 with the text the note paraphrases.
• brain/development/release.md and brain/plans/index.md: the six effect-abort-signal-* files match Abort the request when an Effect operation is interrupted #46. The four new changeset families named here (effect-error-message-*, fetcher-failure-detail-*, dts-private-types-*, effect-abort-signal-*) are the ones Surface the API's error message in ApiError and unit-test the Effect proxy #33, Keep the cause and the response body when a client request fails #37, Export the fetcher types the v0-api and keycloak-api declarations reference #42 and Abort the request when an Effect operation is interrupted #46 introduced.

What I could not check:

• The claim that the generated operation hands config.signal to the client's fetcher. Abort the request when an Effect operation is interrupted #46 does not touch generated code, so this runs on the strength of the unchanged generator, not on this diff.
• The total of 57 changesets. The line it replaces said 34 on 2026-10-15, and the enumerated list already in that section names 42 files by my count of its wildcard entries. The 23 new files named here are consistent with the PRs that added them, but the total needs a directory listing to settle. The PR body says it lists the directory on this commit; a person or the full shepherd should confirm the number and the list before the first release note relies on either.

No secrets, credentials, customer data or personal details appear in the diff.

VERDICT: GO Every statement I could check against the merged #46 and #48 diffs matches, and the two points I could not check are a note for the next release, not a claim this change makes false.

[sferarc-hq]

HQ merge gate on #49: Landing this

• all required checks green (4)
• reviewed by HQ (label "hq-reviewed", which only the reviewer can apply)
• no critical or generated paths in 5 changed file(s)
• no credential material and no new secret references
• 18 changed lines
• review tier: fast path (documentation, notes or comments only). A lightweight review is enough