#549 Manager Daily Digest

[b-at-neu]

Closes #549

SESSION REQUIRED: touches CLAUDE.md — a dispatched agent can't edit it

Summary

• Adds two Vercel Cron routes — daily and weekly manager digests — each a static, single-purpose endpoint rather than one param-dispatched job runner.
• Daily windows since each manager's own last digest (any status; a failed run isn't retried), falling back to 24h for a manager's first-ever digest — gap-safe against a missed/delayed cron fire, and in steady daily operation indistinguishable from a fixed 24h window. Reports new applications per managed position.
• Weekly is a pure reminder, not a recap: it has no time window of its own and reports every application still short of a terminal status (not accepted/rejected/withdrawn/draft) across the manager's positions, as a status-count box per status (colored to match the in-app badge palette), plus their currently-open positions as context. Sent only when that total is greater than zero. This is a design change from the originally-agreed plan (which specified a "new this week" total plus a full status breakdown including terminal decisions) — see the follow-up comment on Manager Daily Digest #549 recording why and what shipped instead.
• Idempotency has no new column: the weekly gate checks for an EmailLog row already existing in the current org week; the daily window is itself the idempotency mechanism (see above).
• Renames EmailTemplateKey.manager_digest (shipped in Email Foundation And Send Log #547, never written) to manager_daily_digest and adds manager_weekly_digest.
• Appends both routes to CLAUDE.md's API-route allowlist (rewritten by Ingest Resend Delivery Events #569) with the bounding clause that they are not a general job runner.

Changes

• prisma/schema.prisma, prisma/migrations/20260911222818_manager_digest_cadences/ — enum rename + add, hand-written ALTER TYPE ... RENAME VALUE / ADD VALUE (data-preserving).
• lib/dates.ts — previousOrgDay, currentOrgWeekStart (org calendar-date arithmetic, DST-safe); previousOrgWeek was added then removed once the weekly digest stopped needing a week-bounded query.
• lib/constants.ts — renamed/added EMAIL_TEMPLATE_VALUES/EMAIL_TEMPLATE_LABELS entries, MANAGER_DIGEST_SEND_SPACING_MS, DAILY_DIGEST_LOOKBACK_MS, reworded complained ("recipient" not "applicant").
• lib/utils.ts — getEmailLogDescription's two bounce sentences reworded to "recipient".
• lib/types.ts — ManagerDigestPosition, DailyDigestRecipient (since: Date), WeeklyDigestStatusCount, WeeklyDigestRecipient (asOfDay: string, unresolved-only statusCounts).
• prisma/data/digests.ts — new; getDailyDigestRecipients fetches raw application rows and buckets per manager by their own since cutoff (can't pre-aggregate until each manager's cutoff is known); getWeeklyDigestRecipients tallies only UNRESOLVED_APPLICATION_STATUSES, no time bound. Both scoped by Position.managers only (never the admin-short-circuiting buildReviewablePositionWhere).
• lib/email/templates.ts — managerDailyDigestEmail (names the exact since-instant), managerWeeklyDigestEmail (table-based colored stat-box grid per status, matching the in-app badge palette), MANAGER_EMAIL_FOOTER.
• lib/email/manager-digests.ts — new; dispatchDailyManagerDigests/dispatchWeeklyManagerDigests, the second named swallow site (per-recipient try/catch, spaced by MANAGER_DIGEST_SEND_SPACING_MS).
• lib/cron.ts — rejectUnauthorizedCron, the bearer-secret check both routes call before any DB access.
• app/api/cron/manager-daily-digest/route.ts, app/api/cron/manager-weekly-digest/route.ts — new.
• vercel.json — the two crons entries.
• CLAUDE.md, docs/ENGINEERING.md, docs/PERMISSIONS.md, docs/WORKFLOWS.md, README.md, .env.example — allowlist, logging swallow site, route-access note, XC-10 workflow entry + PM-8/AD-12 clauses, CRON_SECRET docs.
• tests/unit/dates.test.ts, tests/unit/email-templates.test.ts, tests/db/manager-digests.test.ts — new coverage; tests/db/email-log-queries.test.ts, tests/unit/utils.test.ts — updated for the renamed enum value and reworded copy.

Testing plan

• Set CRON_SECRET in the Vercel preview env and redeploy (also needs RESEND_API_KEY/RESEND_FROM_EMAIL, already set).
• curl -i <preview>/api/cron/manager-daily-digest with no header, then a wrong one → 401 {"error":"Unauthorized"} both times; repeat for …/manager-weekly-digest; /emails shows no new rows.
• As a manager M with two open positions, submit two applications to the first and one to the second.
• curl -H "Authorization: Bearer $CRON_SECRET" <preview>/api/cron/manager-daily-digest → {"sent":1,...}; M gets one email, subject 3 new applications across 2 positions, both positions listed with the right counts, body naming the exact since-instant.
• Click each position link → /manage/applications?positionId=… with the new submissions on top.
• Re-run the same curl immediately → {"sent":0,...,"skipped":1} and no second email.
• Submit one more application, re-run → this time it is included (proves the gap-safe "since last digest" window, not a calendar-day exclusion).
• A second manager N with no new activity gets nothing on the daily run.
• With N having some applications still applied/reviewing (any age) and none accepted/rejected/withdrawn/draft, curl … /api/cron/manager-weekly-digest → one email with a stat box per unresolved status and the open-positions list.
• Re-run the weekly curl → skipped: 1; re-run the daily one → unaffected by the weekly row.
• Give N only terminal-status applications (accepted/rejected) → the weekly run sends nothing to N.
• /emails as an admin → Manager daily digest/Manager weekly digest rows addressed to M/N; both filterable by template; neither appears in any application's Email history.
• Deactivate M, submit another application, re-run both → no digest for M.
• Read both emails in a phone client and Gmail web — rows stay readable at ~320px; each link text reads meaningfully on its own; the weekly stat boxes render (or gracefully degrade to square, undecorated boxes in Outlook).

Automated checks

• npm run prettier:check — pass
• npm run eslint:check — pass
• npm run tsc:check — pass
• npm run test — 781 passed (42 files), run against a local throwaway Postgres container since the worktree's default port 5432 was occupied by an unrelated container
• npm run build — pass; both cron routes appear as dynamic (ƒ) routes

Notes

• Both the daily-window and weekly-content redesigns were direct operator feedback during review, not part of the original agreed plan — see the follow-up comment on Manager Daily Digest #549 for the record of what changed and why.
• Vercel Cron only fires on production deployments, so previews are verified by the manual curls above.
• Cron schedules are UTC, so the local send time shifts an hour across DST (08:00 EDT / 07:00 EST) — deliberate; unaffected by the windowing change, since the daily window is now relative to each manager's own last digest rather than the org calendar.
• A failed digest is not retried — the daily window and weekly gate both key off the last EmailLog row (any status), so a run that errors just gets picked up in full by the next one for the daily digest, or waits a full week for the weekly one; the failed row on /emails is the record either way.
• Genuinely simultaneous calls are the unguarded case; Vercel Cron doesn't overlap runs and the bearer secret bounds who else can call, so no advisory lock was added.
• submittedAt, not ApplicationStatusEvent, is the "received" source for the daily digest — refreshed on a withdrawn → applied resubmission, excludes since-withdrawn rows, unaffected by a reviewer overriding a status back to applied.

[vercel]

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

Project	Deployment	Actions	Updated
aplio	Ready	Preview	Sep 12, 2026 3:28pm UTC

[b-at-neu]

Code Review — Cycle 1 · needs revision

3 open — 3 🟡 Low (see inline)

[b-at-neu]

Revision — Cycle 1

fixed R1-L1, R1-L2, R1-L3 · 9699d22

[b-at-neu]

Code Review — Cycle 2 · needs revision

1 open — 1 🔴 Critical (see inline)

[b-at-neu]

Weekly digest redesign (operator feedback)

Per direct feedback from the operator (not a formal review cycle):

• The weekly email no longer counts "new applications this week." It now sends only when a manager has applications still short of a terminal status (accepted/rejected/withdrawn/draft excluded), regardless of how long they've been outstanding — a pure "you have review work waiting" reminder.
• Status counts now render as colored stat boxes (matching the in-app StatCard/badge palette: blue for applied/reached_out/interview_scheduled, amber for reviewing) instead of a plain text list.
• Subject is now "N applications awaiting your review".
• Confirmed unchanged: weekly still sends Mondays at 13:00 UTC (9am ET); daily digest still windows to the previous org calendar day by design (a submission from "just now" won't appear until the next day's run).
• previousOrgWeek (lib/dates.ts) had no remaining caller after this change and was removed along with its tests.

74284bc on top of the earlier c72e616 fix. All 780 tests pass, prettier/eslint/tsc clean.

[b-at-neu]

Daily digest windowing redesign (operator feedback)

Per direct feedback, the daily digest no longer windows to the previous org calendar day. It now windows since each manager's own last digest (any status, success or failure — a failed run still isn't retried, same philosophy as before), falling back to DAILY_DIGEST_LOOKBACK_MS (24h) for a manager's first-ever digest.

• Gap-safe: a missed or delayed cron fire can no longer drop an application permanently — the next run just looks back further for that manager.
• In steady daily operation this is indistinguishable from the old ~24h window.
• No more same-day exclusion: an application submitted minutes ago now shows up on the very next call, rather than requiring a calendar-day rollover — this was the operator's original testing friction.
• The "gate" is no longer a separate boolean — a manager whose window (since their last digest) yields zero new applications just has an empty digest, whether that's genuinely no activity or a repeat call moments after a successful send. Both collapse into skipped.
• Email copy now names the exact since-instant (formatInstant, e.g. "since Mar 9, 2026, 8:00 AM EST") instead of a single calendar day, since the window can span more than a day after a gap.

0bf327a on top of 74284bc/c72e616. All 781 tests pass, prettier/eslint/tsc clean.

[b-at-neu]

Code Review — Cycle 4 · needs revision

2 open — 1 🟠 Medium, 1 🟡 Low (see inline)

[b-at-neu]

Revision — Cycle 4

fixed R4-M1, R4-L1 · skipped R3-L1 (moot — flagged code removed in 74284bc) · 7e025be

[b-at-neu]

Code Review — Cycle 5 · approved

0 open