quay.io/minio/mc:latest is no longer publicly pullable (401 UNAUTHORIZED),
which broke the Docker publish workflow. Use the official amazon/aws-cli
image to create the bucket idempotently via head-bucket || s3 mb.
* feat(deploy): support Vercel Hobby alongside Docker
* fix(deploy): include PDFKit runtime font assets
* docs(deploy): document Vercel and Docker setup
* docs(deploy): record storage persistence checks
* refactor(deploy): drop scheduled staging cleanup
Staging uploads are deleted after finalization and expired ones are swept
on each new upload, so the Vercel cron job, its route, and CRON_SECRET are
no longer needed. The Deploy with Vercel wizard now asks for two secrets.
* docs(deploy): restructure Vercel guides
Split the Vercel page into a how-to with its environment reference, move the
large RPC staging protocol to an API reference page, and move CI deployment
checks to the contributing section. Point Deploy with Vercel buttons at main.
* chore: remove agent planning records and fix web app description
Delete superpowers plans/specs, ADRs, issue plans, execution briefs, domain
context maps, and Europass research. Describe apps/web as a TanStack Router
SPA served by apps/server.
* refactor(deploy): simplify Vercel support code
- Share one Redis client and key namespace through @reactive-resume/db/redis
for API and auth instead of a second auth-only client.
- Drop the auth seeding retry; the provider already treats concurrent inserts
as no-ops and deployment preparation seeds before runtime.
- Detect staging support from POST /api/storage/stage (404 on Docker) instead
of a separate GET probe.
- Read staged bodies directly; the signed upload already caps their size.
- Close per-subscription Redis connections with disconnect() alone.
- Check Blob health with one list call instead of write/read/delete.
- Remove redundant tsdown onlyBundle list, dead namespace fallbacks, and the
conditional spread in the health status.
* fix(deploy): heal stopped runs with dead owners and keep auth up without Redis
- Run owners refresh a Redis heartbeat until they release their claim. Stop
requests reap the run immediately when the owner has stopped heartbeating,
instead of leaving the thread blocked until the 15-minute TTL reaper.
- Auth and oRPC rate limiters fall back to per-instance memory limits when
Redis errors, instead of rejecting every login or failing requests.
* ci: allow esbuild build for Vercel CLI and register deployment deps with knip
pnpm 12 fails dlx installs with ignored build scripts, so allow esbuild
explicitly. The server bundle keeps @vercel/blob, ioredis, and jose external,
and api/index.mjs is the Vercel Function entry.
* fix(web): send buffered RPC bodies instead of teed streams
Reading a request clone turned the original body into a stream, which
browsers send without inspectable request data and which needs duplex
mode. Send the already buffered Blob for direct requests.
* fix(web): send direct RPC bodies as bytes
Blob request bodies are sent as data pipes, so browser tooling cannot
inspect them. Buffer the original request as an ArrayBuffer and send those
bytes; this restores the e2e save assertions that match on request data.
Opening another dialog during the previous dialog's 300 ms close animation could clear the new dialog and its close handler. This caused the post-merge dashboard lifecycle test to lose the Duplicate Resume dialog after renaming a resume.
- Scope delayed cleanup to the original dialog and require it to remain closed.
- Add regression coverage for both open and closing replacement dialogs; both cases failed before the fix and pass afterward.
- Include the fix in the v5.3.1 release notes.
Validation: `pnpm check`, `pnpm typecheck`, `pnpm test`, and the focused dialog-store suite (12 passing tests).
Prepare v5.3.1 with dashboard search and thumbnail improvements, PDF layout fixes, cover-letter integrations, and self-hosting updates.
- Bump the root version and add release notes with contributor credits, cover-letter REST migration instructions, and the new GHCR image path.
- Align the dashboard authentication plugin with Better Auth's fetch dependency to restore auth-client type inference.
- Regenerate the OpenAPI specification so published validation limits match runtime schemas.
Validation: `pnpm lingui:extract` (no missing translations), `pnpm check`, `pnpm typecheck`, and `pnpm test`.
Dynamic client registration unconditionally rewrote token_endpoint_auth_method
to "none" for every unauthenticated request, downgrading clients that asked for
client_secret_basic or client_secret_post to public clients. Those clients were
issued no client_secret but still authenticated at the token endpoint with the
method they registered, so the exchange failed with 401 invalid_client.
Connecting Composio to the MCP server hit this on every attempt.
Default to "none" only when the client omits the field, which keeps PKCE-only
MCP clients working while confidential clients receive a usable secret.
1.0.9 rejects the Path2D objects pdfjs-dist 6.3.289 passes to fill/clip,
failing every raster test with "Value is none of these types `String`, `Path`".
pnpm reads audit overrides and patch mappings from pnpm-workspace.yaml, which
already carries both. The top-level package.json copies were npm-shaped fields
that pnpm never consults, and they had already drifted: the workspace file maps
'@react-pdf/textkit' unversioned while the package.json copy pinned 7.0.1.
pnpm install --frozen-lockfile still passes with pnpm-lock.yaml unchanged, and
all four patches remain applied at their recorded hashes, which is what shows
the removed block was inert.
Adding packages/pdf to CI turned it red on ubuntu-latest for two reasons,
neither of which is a real regression.
Rasterized pixels depend on the host font rasterizer. The chikorita, ditto
and all-template baselines differ on Linux in rasterSha256 alone: every page
count, item count and text coordinate is byte-identical to the macOS-authored
baseline. Compare the portable geometry on every host and the pixels only on
the platform the PNGs were generated on, so the characterization keeps
protecting layout without asserting another machine's antialiasing.
The picture-fit override case rasterizes twice and timed out at Vitest's 5s
default on a CI runner, with the date suite landing at 3.5-4.7s. Give the
package a 30s timeout rather than leaving every rendering test a runner
slowdown away from failing.
Verified on linux/amd64 in Docker: both files pass, 18/18.
The Lapras section marginTop added in c0c658c0 shifted every date marker in
that template down the page, but the date-layout characterization baseline
was not regenerated, so packages/pdf has been failing on main since. The
delta is geometry-only: same page count (2), same text item count (87), same
markers, x and width unchanged; 17 markers move on y and the raster hashes
follow.
Nothing caught it because the workflow ran test:ci for only server and
tooling, leaving 17 packages uncovered. Replace that filter list with the
full workspace run so a package cannot silently lose coverage again, and
move it after the migration and storage steps that the api suite needs.
Serial execution is deliberate. Running the packages in parallel oversubscribes
the runner and starves the PDF rasterization and API rate-limit suites past
their timeouts; 19/19 pass consistently at --concurrency=1.
Patch and minor bumps across the AI provider SDKs (@ai-sdk/*, ai),
@aws-sdk/client-s3, react-email/@react-email/ui, knip and jszip, with
pnpm-lock.yaml regenerated to match.
Also records the audit overrides and patched dependencies in the root
package.json alongside the existing pnpm-workspace.yaml entries.
The "stops waiting for a slow save while preserving late acknowledgements
and queued edits" test races Playwright's fake clock against real debounce
and network timing, and has failed intermittently on main and in PRs since
it landed. Six prior stabilization attempts, including bumping its timeout
to 60s, did not hold; the latest run on main still exceeded that budget.
The same behavior is covered deterministically with fake timers in
apps/web/src/features/resume/builder/draft.test.ts ("ends a stalled
navigation wait without aborting or discarding the pending save", plus
the queued-edit and pending-snapshot cases), so removing the e2e test
loses no coverage.
Co-Authored-By: Claude Opus 5 (1M context) <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_018z9CKmSSEuS2UFMoqhHWtQ