/prune-memory step D (recall feedback 2026-11-12): BDR-080 title and
lesson narrowed to the V1-1 DEV recette; LRN-080 status implemented,
promoted to rule line .claude/rules/database.md (the rule itself is a
separate /feat in this repo).
Index-only dates and statuses become body lines, frontmatter format: 2,
Index tables removed. memory.sh check rc 0 before and after; digest
statuses unchanged (BLK-008, BLK-009 open).
The suite's supabase-js clients followed redirects by default, so a 3xx
from the sandbox gateway would have carried the apikey header to another
origin. `noRedirectFetch` (`redirect: 'error'`) is passed as `global.fetch`
of both client factories; a regression test with two loopback servers
proves REST, auth and Storage calls stop at the 302. Live proof j and a
second 5/5 run are in the audit log; no production client changes.
scripts/avatar-e2e.sh read the DEV keys through `supabase status`, started
the worker on the DEV network and let the suite write and purge on DEV.
Port the V1-2 sandbox runner (`db-sandbox.sh avatar-e2e`: loopback gateway,
worker in a 600 env file) and the guarded suite (fixed URL, container and
key issuer, per-run accounts); refusals and the positive run proven under
strace in .claude/audits/2026-11-02-avatar-e2e-sandbox-only.md.
The runner read the DEV keys from supabase status and targeted
http://127.0.0.1:54321 (two read-only incidents on 2026-11-01). It now
refuses, before building or any request, every URL but the sandbox gateway
and every key not issued by the sandbox; no DEV URL, port or key remains.
Same bytes as the V1-2 runner; that branch brings the sandbox command.
The card grid had no explicit column below md: its implicit auto track
took the cards' min-content, so the account e-mail never truncated and
pushed the page sideways. One minmax(0,1fr) column; the e-mail wraps.
e2e/shell/responsive.spec.ts sweeps 320-430 px and desktop (RESP-01).
Preflight of DEV recorded; backup adds row fingerprints and Storage bytes; restore check runs on the isolated stack; post-migration SQL, test effects, failure and restore effects; browser acceptance aligned on the owner brief.
014 #237 matched « 20[0-9]{2}- » inside a workspace uuid on a long-lived stack.
The pattern now reads the metadata without workspace_id, still over every
audit row. It would have stopped the DEV procedure at db:test. Found by the
V1-1.6 verifier.
The client journey could fail after it passed: since V1-1.6 the client page
reloads its animals after a restore, and that replay outlived the test. One
support helper unroutes without waiting, for the three shell specs. Found by
the V1-1.6 verifier. Memory: BDR-079, LRN-087 to 089.
Contract §11.8 and §15; D11 and U3 decided, D12 open; SUPABASE.md §3.3 gains
the rehearsed restore check and the owner's browser acceptance on DEV; browser,
test and frontend docs, design README, CLAUDE.md map.
Ten journeys, FR and EN, desktop and 375 px, axe on every state. Seeds go
through quoted SQL literals; db:sandbox e2e-shell passes its arguments to
Playwright.
Routes /app/animals, new, :animalId, edit under one workspace gate; list with
species filter; form (D-R7, U3 birth, no default sex or role, optional first
client); page with current and former relations, end, dates, void; client page
links its animals, adds one pre-associated, links an existing one. Card now
passes aria-labelledby, which it silently dropped.
animals.ts: POST search with species filter and a probe row, read by id with
explicit columns, create with the optional first client in one call, update,
archive. relations.ts: an animal's relations in one select (voided left out),
create, end, dates, void. Kinds duplicate_relation and already_ended added.
Every client screen focuses its heading when it opens (contract §11.7); the
not-found screen did not. Unit and Chromium tests now assert the focus.
Found by the V1-1.5 verifier.
USER_UPDATED (or a retry) set the workspace back to loading, which unmounted
the client routes and lost an open form's typed values. A ready workspace now
stays until the answer; found by the V1-1.5 security audit.
The report's step titles held what the tests typed, the throwaway sandbox
accounts' passwords included, contrary to BROWSER_TESTS.md. The shell suite
keeps the terminal output and failure screenshots; the old report deleted.
Create, search, edit, archive, archived filter, restore; pages of 25;
keyboard only; two accounts; a concurrent edit; a retry after a lost answer
sends the same request id; network failures; the leave guard; a change of
language; FR/EN, desktop and 375 px, axe on every state. signIn shared.
Routes /app/clients, /new, /:clientId, /:clientId/edit behind a workspace
gate that remounts per workspace; debounced POST search, archived toggle and
« Afficher plus » pages; form on field-limits (D-R7) with the phone entry
rule and ISO countries; request id per creation intent; conflict shows the
saved values; archive confirmation with the sole-owner preview; animals
read-only; leave guard on in-app links; Clients entry after the dashboard.
The store reads the personal workspace with the account (ready, loading,
inactive, unconfirmed, missing, error; never created in the browser). A
change of account drops profile, roles, avatar and workspace at once, and a
late answer for the previous account is ignored. generateUuid gives v4 ids on
the plain-HTTP LAN origin, where crypto.randomUUID is missing.
fetchPersonalWorkspaceId, client search (keyset, one probe row), read by id
with explicit columns, create (request id), update (every field), archive,
client animals and archive preview, species labels; RecordFailure kinds of
contract §9.4 plus session_expired and network; RPC answers checked by Zod.
Unit tests and the same functions against real PostgREST on the sandbox.
Contract §8.3 (differences, relation states, caps, lock order across M2 and
M3, concurrency results, migration states, mutations, residuals), helpers
and derived values; SUPABASE.md §3.3 procedure from DEV's current state to
M3 (not run) and §3.5; TEST_STRATEGY, plan note, backend table, CLAUDE.md.
A failing stage prints FAILED with its name and exits non-zero; e2e:shell is
reported NOT RUN (it needs the isolated stack); both suites keep their own
commands.