Skill: Daily Brief

When to activate

Scheduled daily task (e.g. end of day) → daily mode.
Scheduled weekly task (e.g. Monday morning) → weekly mode (synthesis of the previous week).
Manual invocation: "daily brief", "recap of the day", "what happened", "weekly review".

Preflight

Read _CLAUDE.md (rules, structure, conventions) and 00-inbox/MY-PROFILE.md.
Get the real timestamp.
Determine the mode: daily (default) or weekly (Monday morning or on request).
Define the time window:
- daily: today 00:00 → now.
- weekly: Monday of last week 00:00 → Sunday of last week 23:59.

Process

Step 1 — Multi-source collection

Collect from whatever sources are connected (via available tools / MCP servers). Run in parallel where possible.

Primary vs secondary sources. Read primary-communication-channels from 00-inbox/MY-PROFILE.md. Anything listed there (email, slack, teams, calendar, …) is a co-primary source: treat it as a peer of email and calendar in volume of attention, in collection breadth, and in the brief structure. If slack (or similar) is listed, do NOT under-weight it just because it's high-volume — chat is where most of the day's signal lives for many users.

1.1 Email

Emails received in the window.
Filter out: newsletters, automated notifications, marketing.
Keep: conversations with colleagues, relevant externals, decision threads.
Extract: subject, sender, one-line summary, implicit urgency.

1.2 Calendar

Events in the window (past + ongoing).
For each: title, participants, duration, notes/description.
Cross-reference with 04-meetings/: if already ingested → link, else → flag for potential ingestion.

1.3 Drive / files

Docs modified or shared in the window.
Filter out: your own minor edits.
Keep: new docs received, collaborative docs touched by others, meeting transcripts created.

1.4 Chat (Slack / Teams / etc.) — co-primary when listed in `primary-communication-channels`

Collect broadly when chat is a primary channel; collect narrowly otherwise.

Always (whatever the channel's status):

DMs received in the window.
DMs sent by the user (signal of what the user cared enough to write).
Mentions of the user in any channel.
Replies in threads the user has participated in (started or replied to) within the window.

Additionally when chat is a primary channel:

Activity in any channel the user actively follows — defined as: channels where the user has posted, replied, or reacted within the last 14 days, even if not in priority-chat-channels. This catches the long tail of where the user actually works.
Reactions the user placed on others' messages (signal of interest / agreement).
Saved / bookmarked messages.
Huddles or calls in chat (if surfaceable).

Filter out (both modes):

Bot notifications, CI noise, build-status messages.
Emoji-only reactions on bot messages.
Off-topic channels explicitly excluded in MY-PROFILE.md (chat-channels-to-ignore).

Extract per channel/thread: topic (one line), participants, message count, the user's own contribution, any decision/commitment, any unresolved question.

Volume guardrail. A busy Slack day can swamp the rest. Aggregate by topic, not by message: cluster messages across channels/threads into themes (e.g. "onboarding script friction" pulled from #onboarding, #cs-team, and a DM with Alex). One theme = one bullet, with the threads cited as sources. Cap at ~10 themes; the synthesis in Step 3 will reduce further.

1.5 Internal vault

Notes created or modified in the window (all sections except 07-archive).
Especially: new decisions, ingested meetings, people updates, unprocessed braindumps.

Step 2 — Triage and prioritization

Categorize the collected items:

2.1 Decisions / Commitments

Any action the user committed to ("I'll handle it", "ok I'll do it", an assigned action item).
Decisions made by others that impact the user's projects.

2.2 Weak signals

A topic mentioned repeatedly across channels.
Detectable interpersonal tensions.
Shifts in tone/priority from stakeholders.
First mention of something that could grow.

2.3 Awaited follow-ups

Action items the user owns with an approaching deadline.
People waiting on a reply from the user.
Past commitments to verify.

2.4 Learning / inputs

Articles shared, docs to read.
New concepts/frameworks mentioned.
Best practices surfaced by teams.

2.5 People intel

People interacted with today.
New information about known stakeholders (role change, etc.).
New people encountered (candidates for a person note).

Step 3 — Intelligent synthesis

⚠️ Major anti-pattern: do not produce a flat exhaustive list. The brief must be actionable and synthetic.

Synthesis rules:

Top 3–5 topics only as main sections (not 15) — these synthesize across all sources.
One sentence = one signal (no vague paraphrase).
Always link to the source (email, meeting, doc, vault note, chat thread).
Recency markers for external claims.
## Themes from Slack is separate from the top-5 when chat is a primary channel. Chat themes do NOT compete with email/meeting topics for top-5 slots; they get their own section so high-volume Slack signal isn't drowned out (see Step 4 template).

Step 4 — Generate the note

Daily path: 01-daily/YYYY-MM-DD.md Weekly path: 01-daily/YYYY-WW-weekly.md (e.g. 2026-21-weekly.md)

⚠️ If the daily note already exists (because other skills wrote into it during the day): APPEND, don't replace. Add the brief at the top under ## Daily brief, keeping the existing sections (braindumps, ingested meetings).

Daily format:

---
date: YYYY-MM-DD
type: daily
tags: [daily]
ai-first: true
---

## For future Claude

Daily brief for [date]. Generated automatically at [timestamp] aggregating email,
calendar, drive, chat, and internal vault activity. Top themes: [3-5 keywords].
[Note any unusual signal or urgent follow-up.]

## TL;DR (3 lines)

1. [The most important thing today, one sentence]
2. [The second thing, one sentence]
3. [The third thing, one sentence]

## Decisions & commitments

- [Committed action] — source: [[email/meeting/chat]] — deadline: YYYY-MM-DD
- [Decision by X impacting project Y] — source: [...]

## Decisions reversed — pending postmortem

> Render only if Step 6.7 detected one or more decisions flipped to `status: reversed` this
> window. Otherwise omit the section entirely.

- [[05-decisions/<slug>]] — reversed YYYY-MM-DD.
  - Hypothesis: <one sentence>
  - Distilled lesson: <one sentence>
  - Vault references found: N decisions, N wikis, N lessons (manifest in [[06-knowledge/<slug>-postmortem]])
  - Proposed updates: ready for your review — accept all / cherry-pick / dismiss.

## Top topics of the day

### [Topic 1]
[2-3 sentences synthesizing what happened on this topic, with linked sources]
**Sources**: [[04-meetings/...]], email from [[02-people/...]], chat thread #channel
**Next**: [action or open question]

### [Topic 2]
[...]

## Themes from Slack

> Only emit this section when `slack` (or another chat tool) is listed in
> `primary-communication-channels`. Aggregate by **theme**, not by message; one bullet per
> theme, citing the channels/threads it spans. This section is independent of the top-5 cap.

- **[Theme 1]** — observed in #channel-a, #channel-b, DM with [[02-people/...]] — N messages,
  the user participated in M. Signal: [one sentence].
- **[Theme 2]** — [...]

(If the day was light on chat: a one-liner like "Slack quiet today" is fine.)

## Weak signals to dig into

- [Signal 1] — observed in [sources] — hypothesis: [...]
- [Signal 2] — [...]

## Pending follow-ups

[Fed from the Overdue + Today buckets of [[TODO]]. Each links back to its source note.
These lines are MIRRORS of existing tasks — they carry the source anchor in the wikilink
and never get a new `^t-id` of their own (task-roundup reconciles through the referenced
anchor).]
- [ ] [Action] — [[TODO]] → [[<source-note>#^t-id]] — due YYYY-MM-DD
- [ ] [Action] — [[<source-note>#^t-id]] — due YYYY-MM-DD

## People touched today

- [[02-people/Alex Rivera]] — 1-1 + 2 emails — topic: training
- [[02-people/...]] — [...]

## Inputs of the day

- [Article/doc shared by X] — source: [link or drive path] — relevance: [why]

## Braindumps of the day

[Links to braindumps created via the braindump skill — that skill appends here]

## Meetings ingested today

[Links to ingested meetings — meeting-ingest appends here]

## Docs ingested today

[Links to ingested documents — doc-ingest appends here]

Weekly mode also runs knowledge-build (two phases):

Lessons sweep — conservative pass that proposes new/updated 06-knowledge/ lesson notes flagged needs-review: true. List them in the review under ## Knowledge updates.

Curator sweep (Mode C.2 of knowledge-build) — rebuilds auto-maintained hub listings, refreshes _INDEX.md, flags orphans / near-duplicates / aging stubs / stale wikis, and proposes new hubs when 3+ notes cluster on an unhubbed domain. Structural changes always preview and ask. Surface the health counters and any pending proposals at the top of the weekly review under ## Knowledge garden.

Weekly format (different — synthesis-focused vs. collection):

---
date: YYYY-MM-DD              # first day of the analyzed week
week: YYYY-WW
type: weekly-review
tags: [weekly]
ai-first: true
---

## For future Claude

Weekly review for week [WW] of [YYYY]. Covers [start date] → [end date]. Compiled
from 7 daily briefs + a direct vault scan.

## Knowledge garden

[Weekly mode only — output of the curator sweep: health counters, passes to stabilize,
pending proposals (new hubs, merges, stubs aging out). Omit if the sweep didn't run.]

## Knowledge updates

[Weekly mode only — lessons proposed/updated by the lessons sweep, flagged
`needs-review: true`, for you to confirm.]

## Themes of the week

[3-5 cross-cutting themes that emerged, with backlinks to daily notes / meetings / decisions]

### [Theme 1]
- Appeared: [first signal]
- Evolution: [how it moved]
- Current state: [resolved / open / escalated]
- Linked to: [[03-projects/...]]

## Decisions of the week

[Structured recap of important decisions, with links to 05-decisions/]

## People focus

[Who emerged most this week? Why?]

## Patterns detected

[Cross-day recurrences, confirmed weak signals]

## Anti-patterns / Friction

[What isn't working, what recurs negatively]

## Energy & focus

[Honest synthesis: where did the energy go, is it aligned with declared priorities?]

## Implicit plan for next week

[What the daily notes suggest as priorities for the coming week]

## Stubs to process

[Incomplete notes or notes awaiting clarification]

Step 5 — Meeting auto-orchestration (CRITICAL)

⚠️ This is what makes the system run without manual intervention.

For each meeting transcript detected during the window:

5.1 Whitelist check

A transcript is auto-ingestable ONLY if ALL conditions hold:

✅ Matches a calendar event in the window (by title, participants, or time).
✅ Calendar participants ≥ 2 (including the user).
✅ Calendar duration ≥ 15 minutes.
✅ NOT in the sensitive-meetings list of 00-inbox/MY-PROFILE.md.
✅ Not already ingested (cross-check 04-meetings/ by date + slug).

5.2 Auto-ingest (for transcripts that pass the whitelist)

Invoke meeting-ingest in auto mode, with these differences vs. manual:

Frontmatter: ingestion-mode: auto and needs-review: true.
"For future Claude" mentions: *Auto-ingested via daily-brief on YYYY-MM-DD. needs-review flag is true until the user confirms.*
People propagation handled in step 6 (don't duplicate here).
Hard-to-reverse decisions detected → flag needs-validation: true rather than creating directly in 05-decisions/.

Cap at 5 auto-ingested meetings per daily-brief run to avoid saturating context. If > 5, ingest the 5 longest/most important and list the rest for the next day.

5.3 Flag for human validation

For transcripts that do NOT pass the whitelist:

Mention explicitly in the brief:

⏸ Awaiting manual validation:
- 1-1 with [manager] (sensitive, transcript available) → want me to ingest it?
- Orphan transcript (no calendar match) → ignore or ingest with context?

Auto-ingest NOTHING.
The user validates yes/no the next day when reading the brief.

Step 6 — People auto-update (CRITICAL)

For each person the user interacted with during the window (detected via calendar/email/chat):

6.1 Interaction detection

For each person:

Calendar events (participants including the person).
Email threads (sender OR recipient = this person; exclude generic cc).
Chat DMs (direct DM or mention in a critical channel).

6.2 Fuzzy-match against 02-people/

If the note exists → proceed to update (step 6.3).
If the note does NOT exist → create NOTHING automatically. Append to the "New people detected" section of the brief for the user to validate next day.

6.3 Update the note (auto-logged)

For each person whose note exists:

Frontmatter:

Update last-interaction: YYYY-MM-DD (today).
Update updated: YYYY-MM-DD (today).
Clear staleness-flag if present (self-healing).

Timeline: Append ONE aggregated entry for the day:

### YYYY-MM-DD — Daily interactions (auto-logged)
- Calendar: [events with this person, format "1-1 30min 'title'"]
- Email: [N threads, main topic if detectable]
- Chat: [N DMs or mentions, channel if relevant]
- Source: daily-brief auto-propagation

⚠️ No interpretive observation. Raw facts only. The "Compiled truth" section is NEVER touched in auto mode.

⚠️ Deduplication: if an (auto-logged) timeline entry already exists for the day (rare: daily-brief run twice), update the existing entry instead of creating a new one.

6.4 Quantity cap

Auto-update up to 20 people notes per run. If > 20 detected (rare: very busy day), prioritize:

Direct reports
Manager
Recurring peers
Critical stakeholders (per MY-PROFILE.md)
Others

List the non-updated ones in the brief: "12 other people interacted with today not auto-updated (volume cap). Invoke people-update manually if needed."

Step 6.5 — Task roundup & reconciliation (CRITICAL)

Run the task-roundup procedure (see .claude/skills/task-roundup/SKILL.md):

Collect the action items you own from today's new/updated notes (ingested meetings, decisions, follow-ups) plus anything still open across the vault.
Assign a ^t-id block-ID to any new action line that lacks one (additive edit).
Reconcile checkboxes both ways with TODO.md at the vault root: a box checked in TODO.md since the last run flips its source line to [x] ✅ <date>, and vice versa.
Refresh TODO.md, bucketed by due date (overdue / today / upcoming / later / no date / waiting-on-others / done).

⚠️ Completing a task in 02-people/ or 05-decisions/ is a checkbox toggle + ✅ <date> stamp only — never touch Compiled truth in auto mode.

Then surface the result at the top of the brief: feed ## Pending follow-ups from the Overdue and Today buckets (with the [[TODO]] backlinks), and flag the overdue count.

Step 6.7 — Reversed-decision postmortem auto-trigger (since v4.0)

Scan 05-decisions/ for any note whose status: changed to reversed since the last brief (compare against the previous daily/weekly's updated: window). For each detected flip:

Invoke challenge-decision in postmortem mode for that decision (auto). The skill runs Steps P1 and P2 fully — extracts the hypothesis, cross-references the vault — and produces the previewed edits for Step P3.
Postmortem auto-mode never applies the cross-vault edits. It surfaces them as a manifest in this brief under a new section ## Decisions reversed — pending postmortem, with the proposed wiki/lesson updates listed inline for human approval.
The postmortem analysis note (06-knowledge/<decision-slug>-postmortem.md or appended to the reversed decision) is created with needs-review: true.

If no decision flipped since the last brief, skip this step silently (don't render an empty section).

Detection rule of thumb (auto-conservative):

A decision is "flipped this window" if its file's git history shows the status: field changed to reversed between the previous brief's window-end and now, or if a timeline entry titled ### YYYY-MM-DD — Reversed was appended within the window.
Don't re-trigger on decisions already reversed before the last brief — those were already handled (or skipped) previously. The postmortem fires once per flip.

Step 7 — Propagation

Link from the previous daily: append "→ Next: [[YYYY-MM-DD]]" in the previous day's daily note.
For the weekly: link from the 7 daily notes of the week to the weekly review.

Step 8 — Report

✓ Daily brief generated: 01-daily/2026-05-22.md
✓ Sources analyzed: Email (12), Calendar (4 events), Drive (3 docs), Chat — primary (47 msgs across 9 channels, clustered into 5 themes)

📥 AUTO-INGESTION
✓ 3 meetings auto-ingested (needs-review: true):
  - [[04-meetings/2026-05-22-team-sync]] (15-min morning review recommended)
  - [[04-meetings/2026-05-22-stakeholder-onboarding]]
  - [[04-meetings/2026-05-22-project-sync]]
✓ 8 people notes auto-updated (last-interaction + auto-logged timeline)
✓ 2 staleness-flags cleared (Alex, Jordan)

⏸ AWAITING VALIDATION
- 1-1 with [direct manager] (sensitive, transcript available) → want me to ingest it tomorrow?
- 2 new people detected (never seen in the vault):
  - "Riley Chen" (3 emails today, seems to be a PM) → create a note?
  - "Sam Patel" (1 meeting + 1 email, external context) → create a note?

✅ TASKS
✓ TODO.md refreshed — ⏰ 2 overdue · 📅 3 due today · ⏳ 5 waiting on others
✓ Reconciled 3 check-offs with source notes (2 done here, 1 done in source)

🎯 SYNTHESIS
✓ Top topic: onboarding script friction
✓ Weak signals: 2 (to dig into)
⚠️ 3 follow-up deadlines approaching (D+2, D+3, D+5)

The vault updated itself.

Configuration (from MY-PROFILE.md)

Read these from 00-inbox/MY-PROFILE.md:

primary-communication-channels — sources to treat as co-primary with email/calendar (e.g. [slack], [teams, slack]). When set, the matching Step 1 collector goes into broad mode and the brief renders the corresponding dedicated section (## Themes from Slack).
Priority chat channels — a narrower whitelist within the chat source for must-read channels; still useful as a hint, but no longer the gate for whether chat is collected.
chat-channels-to-ignore — explicit blocklist for noisy/off-topic channels.
Critical stakeholders — prioritize in the "People touched" section.
Active projects (status: active in 03-projects/*) — weight related signals.
Brief time / weekly time — when the scheduled runs happen.
Tone — direct, no "great progress today!" filler.
Top topics cap — default 5 (does NOT include ## Themes from Slack).

Anti-patterns to avoid

❌ Flat exhaustive list — synthesize, don't dump. ❌ Re-paraphrasing what's already in the vault — link, don't duplicate. ❌ Inventing themes — if the day is quiet, the brief is short. That's fine. ❌ Corporate tone — direct and concrete. ❌ Self-congratulation — no saccharine "great progress!". ❌ Skipping weak signals — that's often where the value hides. ❌ Overwriting an existing daily — append, never overwrite.

Special cases

Quiet day

If the day is quiet (no meetings, few emails, weekend): a minimal 5–10 line brief. No need to fill space.

Very busy day

Beyond 5 topics, don't overflow — keep the top 5, mention "+ N minor topics (see vault)".

Day off

If the user marked OOO in calendar: skip the brief, or generate only a "follow-ups awaiting your return" section.

First use (empty vault)

Degraded brief: just the external sources, no vault cross-referencing. Mention that value grows as the vault fills.