daily-brief

name	daily-brief
description	A short, scannable morning briefing — today's calendar, priority unread, follow-ups going cold, and what needs attention. Activate when the user says "/assistant:brief", "morning briefing", "what's my day look like", or on a morning schedule.

Daily brief

One screen the user reads with their coffee that tells them exactly what today needs. This is the anchor of the scheduled-task setup — most users run it weekday mornings. Catalog entry: job_id: morning-briefing in scheduled/schedules.yaml.

Read the profile first

Load working hours from {assistantPath}/policies/calendar.policy.md, VIP tiers from {assistantPath}/policies/email.policy.md, and voice from the profile (the brief is written to the user, in their preferred style — tight, scannable, no filler).

Schedule health check (interactive runs only)

Before assembling the brief on an interactive run (/assistant:brief), check for a missed scheduled morning briefing. Do not run this block when the current session is the scheduled run recovering the job.

1. Load {working-folder}/scheduled/morning-briefing.yaml if present. If missing, treat as "never scheduled" — suggest /assistant:schedules once, then proceed. If corrupt, note "health ledger unreadable — re-run /assistant:schedules" and proceed.
2. Read the morning-briefing entry. Skip miss detection when surface is managed or cloud-code — local miss logic does not apply.
3. On surface: local, on a weekday, when local time is past the expected run window (cron fire time + 90 minutes; default 08:00 + 90m = 09:30), detect a miss when either:
   • brief-{today}.md is absent in the working folder, or
   • last_run_at is stale (before today's expected window).
4. On miss: increment miss_count_7d in the job file, persist it, then append one health block after normal brief output (see below). Do not auto-run a catch-up brief — the user may be intentionally running early or late.
5. When miss_count_7d >= 2 for this critical job on local, add the escalation line to the health block.

Expected run window: cron from catalog (0 8 * * 1-5) + 90 minutes. Use profile working_hours.timezone when set; otherwise the user's stated local timezone.

Schedule health block (append when miss detected)

### Schedule health
- **Missed:** Morning briefing (expected ~08:00 weekdays on *local* surface)
- **Last success:** YYYY-MM-DD — `brief-YYYY-MM-DD.md`
- **Suggestion:** If this happens more than once a week, move this job to the managed-agent cookbook (`managed-agents/morning-briefing/agent.yaml`) or a Claude Code cloud schedule. Run `/assistant:schedules` to walk the decision tree.

When miss_count_7d >= 2, add:

This job has missed twice in the last seven days on a local schedule. Consider moving only this job to managed-agent or cloud-code — not your whole setup.

Max one health block per chat turn.

What it assembles

Pull from whatever is connected; degrade gracefully when something's missing.

1. Today's calendar — meetings with times and attendees; flag anything external or VIP; note gaps and any conflict/no-buffer issues. When back-to-back meetings or missing buffers are detected, append one line: "Run /assistant:calendar protect to draft buffer blocks." When habits are configured in calendar.policy.md and a pasted agenda is already loaded for the calendar section, optionally footnote: habit: N gaps this week (cheap scan only — no extra connector round-trip). Optionally footnote: calendar: N buffer gaps. Hand scheduling detail to calendar-scheduling protect mode when the user confirms — default is offer, not silent inline scan (keeps brief under ~400 words).
2. Priority unread — prefer today's sweep artefact over a live re-triage (see Sweep handoff below). Fall back to ~~email/~~chat top items or pasted unread when no recent sweep exists. VIP-first; only what matters before noon.
3. Follow-ups going cold — from follow-up-tracking: what's silent past threshold.
4. Due tasks — from TASKS.md: due or overdue today (via task-management).
5. Needs attention before noon — the 1–3 things that will hurt if missed. Merge VIP/needs-reply lines from the sweep artefact when present.

Sweep handoff

Before scanning live mail for priority unread:

1. Look for sweep-YYYY-MM-DD-{slot}.md files in the working folder for today's date (any slot: 0800, 1200, 1600).
2. Use the most recent sweep file if its timestamp is within the last 12 hours (tolerates the 08:00 sweep/brief race — brief may run before sweep finishes).
3. Pull VIP and needs-reply lines from that file into Before noon / Priority unread — do not re-run full inbox triage when a fresh sweep exists.
4. If no sweep file within 12h, note one line: No sweep artefact since {last sweep date or "today"} — using live unread. Then use ~~email or ask for paste.

Sweep files are written by scheduled /assistant:inbox triage (08:00) and /assistant:inbox sweep (12:00/16:00) per skills/inbox-triage/SKILL.md. The 08:00 triage run writes the same sweep-YYYY-MM-DD-0800.md artefact shape the brief consumes.

Ledger out-of-sync fallback

When loading sweep context or checking inbox schedule health:

1. If {working-folder}/scheduled/inbox-triage-am.yaml or inbox-sweep.yaml has last_run_at: null (or missing) but a matching sweep-YYYY-MM-DD-{slot}.md exists for today, treat the artefact file's modification time as the implied last run.
2. Flag ledger out of sync in the brief or health output — do not report "no prior sweep on record."
3. Suggest re-running the ledger command from skills/inbox-triage/SKILL.md or the next scheduled job to repair the heartbeat.

Output shape

Keep it under ~400 words. Lead with the day's shape.

Morning brief — Tue 8 Jul

Today: 3 meetings, first at 10:00.
  10:00 Acme quarterly (Sam Lee, external) — deck due, decision on Q3 scope
  14:00 1:1 Dana
  16:00 Focus block (protected)

Before noon
  • Reply to Priya re board deck (VIP) — draft ready
  • Send Sam the revised deck ahead of 10:00

Going cold
  • Acme contract — silent 4 days, nudge drafted

Due today
  • Submit expense report

Quiet otherwise. Have a good one.

Scheduled use

When run as a scheduled task, save output to brief-YYYY-MM-DD.md in the working folder so there's a history. See skills/schedule-setup/SKILL.md for the packaged 8am prompt.

Heartbeat: at end of a scheduled run, update the ledger via bash (required):

python3 scripts/update_ledger.py \
  --job-id morning-briefing \
  --status {success|partial|failed} \
  --artifact brief-{today}.md \
  --working-folder {working-folder}

Before ending, confirm last_run_at in scheduled/morning-briefing.yaml matches this run.

Standalone

No connectors? Ask the user to paste their calendar and top unread, or brief from TASKS.md and memory alone. Still useful.

Rules

• Read-only and draft-only: the brief surfaces and drafts, never sends or books.
• Short beats complete. If it doesn't change what the user does today, leave it out.

Approval frame

Follow rules/approval-frame.md when surfacing items needing attention — especially cold threads and draft-ready VIP items.

Use all four sections when the brief includes reviewable work. When a draft-ready item has a queue id, you may footnote the id in brief prose (e.g. "draft ready — rq-2026-07-08-reply-priya-deck").

Do not write queue items for read-only calendar or FYI sections unless a separate reviewable artefact exists.