feat(front-2): full design-parity handoff — Gray UI stack, all surfaces, testing pass - #806
Merged
Merged
Conversation
…pe fixed table layout
…eviation SPEC's 2b/3a grids assume columns the API does not return (Role/Profiles/2FA/ Last-active; Plan/Owner/Created). Profiles and invitations use the literal SPEC pixel values; staff-users and tenants keep the widths that have a SPEC counterpart and let the longest free-text column absorb the remainder. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…eparator A single '*' in a Playwright glob compiles to ([^/]*) and stops at '/'; only '**' becomes (.*). A trailing single star therefore matches the collection endpoint but never its sub-paths, so the handler is dead code and the request escapes to the real API while the spec still passes. This silently defeated three specs across two packets despite being documented in the trap list. Scan e2e/ (previously unscanned) and add no-single-star-route-glob. Existing sites are annotated with a reason rather than widened, because widening changes what a mock intercepts and must be argued per spec. Suppression requires a reason after the rule id, so it cannot be added thoughtlessly. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
… paragraph Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
'demo-user-id' is not a GUID, so the API answers 400 malformed-id and the route renders the error view — which has no .publy-detail-grid. The other shell specs get away with that id because they only assert on the panel. The assertions were already right; only the navigation was wrong. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…eedback batch Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
P13 sized the icon container but the svg glyph kept its call-site size-7 (28px), which wins over a layered rule. An un-layered .publy-state-icon svg rule (100% of the 40/48px box) beats every Tailwind layer, valorizing the icon as intended without editing 61 call sites. Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>
…op a fabricated chip
…tations back-link reloading
…earch 50 worktrees under .worktrees/ (45 with their own node_modules) were being watched with no excludes configured, ballooning VS Code's remote fileWatcher process to ~4.4GB RSS. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
files.exclude hides folders from the tree entirely, not just from search and the watcher — that broke browsing into worktrees and scratch notes. They stay out of files.watcherExclude/search.exclude, so no RAM regression. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com>
…ale to error views, arrow back-links
…ctive actions out of headers, consistent back-links
…ions guide Captures the standing decisions from the front-2 parity review (borders, radius, destructive-action placement, error/empty states, malformed-id -> 404, back-links, data honesty, navigation) so future work inherits them by default. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
…default The default-size (36px) button radius is 12px under the owner-ratified per-size scale (xs 8 / sm 10 / default 12 / lg 14). The p15 radius change updated the artboard-assertions but missed this e2e pin; the button renders correctly. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
Two files under staff-users/$userId/ were unformatted since wave 5 (71296ed); the CI 'Check formatting' step (pnpm format = oxfmt --check) fails on them. Formatting-only (line-wrapping); typecheck 0, affected tests 22/22. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
This was referenced Jul 15, 2026
Closed
.dump/orchestration/W4-API-report.md and W4-API.STATUS were force-added past the .dump gitignore in wave 4; they are orchestration scratch and must not land on develop. No product code affected. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>
radandevist
added a commit
that referenced
this pull request
Jul 17, 2026
## Why Epic A (#194) had no authoritative design for the jobs/worker infrastructure. #806 shipped a typed invitation-email outbox; #632 asked for the generic design that supersedes it. This document is that reference: `APP_ROLE` hosting, a single generic `job_queue`, the email fold, and the build order for #633–#636. ## What this is `docs/implementation-plans/jobs-worker-infrastructure.md` — the authoritative architecture reference for PublyApp's background jobs and worker. Ratified decisions: - **D1** — role-based hosting: one codebase, one image, `APP_ROLE=api|worker|all`. - **D2** (owner revision) — emails are **jobs on the generic `job_queue`**; delivery history lives in `email_log` from day one. Supersedes the two-lane/typed-outbox ruling; `invitation_email_outbox` folds via expand/contract. - **D3** — pure Postgres, no broker. ## How it was reviewed Authored, then put through an **adversarial merge-challenge loop** against GPT-5.6-sol at high effort, on the owner's instruction. **10 rounds**, ~80 findings, remediated by Claude Opus 4.8 (cross-family reviewer discipline). Findings by round: 17 → 12 → 9 → 7 → 3 → 5 → 3 → 3 → 4 → 7. All 10 review records are committed under `docs/reviews/jobs-infra-design-challenge/` — the full adversarial history, not a summary. The loop caught design defects that would otherwise have shipped, including: unfenced stale schedule triggers surviving a cron edit (`schedule_epoch`); an expiry sweep that destroyed the evidence of its own expiry; a retention job that deleted the row making another retention job safe; and repeated cases of a claimed property being stronger than the mechanism enforcing it. ## Status — read this before treating any section as settled **The loop did NOT end in `CHALLENGE: MERGE-READY`.** It ended by owner decision after round 10, with round 10's remediation **unvalidated** — every prior round was checked by the next, and that repeatedly mattered (round 8's headline fix graded *Mis-absorbed*; round 9 found two blockers inside round 8's new mechanisms). The status block and §11 record this honestly. §11 **"Known open items"** lists what round 10 could not close, each filed as a GitHub issue against the phase that will implement it. **O6–O31** are author-decided and remain pending owner ratification. A five-round-old fabrication was found during the final round and corrected: the `audit_logs` INSERT promised for engine transitions named four columns the shipped table does not have, against a `user_id NOT NULL` FK the engine cannot satisfy. Every "immutable AuditLog entry" specified for an engine transition since round 5 was unbuildable. Engine evidence now writes `job_dead_letter_events` (O30). ## Doc-vs-code The code on #853/#854/#857/#858 implements an earlier state of this design. The document is deliberately ahead of it and marks the gaps inline as captain-alignment items; a gap analysis against this final version follows separately. Where doc and code disagree, **the doc is the intent and the code is the backlog** — except where an item is marked a suspected live defect, which is a bug. ## Verification Doc-only change. No code, schema, or test touched. Cross-references resolved, fences balanced, review records byte-identical to their sources. Closes #632
This was referenced Aug 22, 2026
radandevist
added a commit
that referenced
this pull request
Aug 25, 2026
…#823) The static class-string simulator (~900 lines) could not model @layer precedence, so it kept certifying a ring that a non-layered rule had already beaten in every real browser — the exact blindness four reviewers converged on during the #806 rounds 6 & 7. What stays under vitest (no cascade model needed): - --publy-focus-ring >= 3:1 against surface tokens, light AND dark themes - ui-F1 choice-chip pre-fix regression proof - fail-closed tripwires (inline ring styles, runtime string composition) - coverage tripwires (button/badge variant keys pinned, routes/ consumer discovery) The rendered-cascade half of the contract moves to the Playwright spec added in the next commit.
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Closes #805.
The complete front-2 design-parity handoff: 159 commits, 325 files, +54k/−5.4k. Supersedes the 5-PR stack plan (#804 closed unmerged); lands as one unit. Every round below was gate-verified green when it closed (build + vitest + full docker-Playwright e2e); the branch tip stands at 156/156 e2e, 693/693 vitest.
Act 1 — Design-system foundation
Gray UI token layer (
--publy-*, both themes), design-invariant guard tests, route metadata, rebuilt authenticated shell, product UI primitives, restyled data-table foundation; staff-admin tokens + shell (handoff tasks 1–2).Act 2 — The Gray UI pivot
81e5a70f): HeroUI removed entirely; ~27 in-house primitives insrc/components/ui/on@base-ui/react+ CVA,@tabler/icons-react; every route family migrated; design-system guard script wired intopnpm test.141475e3…): nav-registry decision, design guards, app shell, table + overlay/form foundations; all 7 latent e2e failures cleared (a054182f).Act 3 — Slices, enrichment, testing pass
ssr:falseauthed layout had anullpending fallback, now ships a static shell skeleton in the SSR HTML itself (cold-boot-stability.spec.tsguards it).Deploy prerequisite
Uploads persist to local disk — production needs a Dokploy persistent volume at
/app/.artifacts/storagebefore this deploys, or uploaded logos vanish on redeploy.Merge-strategy note
A squash collapses 159 commits into one on develop. If you want the per-round history browsable from develop, use a merge commit for this PR instead of the usual squash.
🤖 Generated with Claude Code