daily-summary

name	daily-summary
description	Use when drafting the end-of-day summary. Reads today's meetings, Slack, and memory files to append a structured summary to today's daily note.

Daily Summary

Synthesize today's activity into a structured summary and append it to today's daily note.

Step 0 — Determine paths

Read memory/me.md to get:

• Workspace parent folder
• Daily notes path (e.g. {{WORKSPACE}}/daily-notes/YYYY/MM/YYYY-MM-DD.md)
• Meeting notes path (e.g. {{WORKSPACE}}/meeting-notes/YYYY-MM-DD/)
• Memory path

Use currentDate from context if available, otherwise infer from file timestamps.

Previous working day: step back one calendar day, skipping weekends.

---

Phase 1 — Grounding (run first, before anything else)

Run a single grounding pass. Do not start Phase 2 until complete.

Tasks:

1. Meeting notes: Glob {{MEETING_NOTES_PATH}}/YYYY-MM-DD/*.md. If the folder doesn't exist, note that and continue. For each file:

   • Read in full (use offset/limit of 150 lines if needed)
   • Extract: key decisions, open questions, your explicit action items or commitments
   • Extract from YAML frontmatter: created field HH:MM — check memory/me.md for timezone; if using Granola, timestamps are UTC and must be converted to local time
   • Note filename without .md (for wikilink) and derive display name: replace _ → /, then _ before space or end-of-string → :

2. Carry-forward to-dos: Read the last 5 working days' daily notes. For each, find ### To-dos inside ## Daily Summary and extract:

   • Unchecked - [ ] items
   • Checked - [x] items (to avoid regenerating completed work)

3. Context: Read memory/projects.md, memory/teams.md, memory/people-signal.md, and skim other memory files. Note active projects, key people, today's meeting themes, and initial to-do carry-forward candidates.

Output of Phase 1: Meeting summaries with times and action items, to-do carry-forward candidates (checked/unchecked), and a compact project/team context summary. Passed as context to all Phase 2 agents.

---

Phase 2 — Signals (run in parallel, after Phase 1 completes)

Dispatch subagents simultaneously. Each receives Phase 1 output as context. All return findings only — no file writes.

---

Subagent A — Slack

Read memory/me.md for your Slack user ID. Read memory/slack-channels.md for the full channel list with IDs and tiers. Do not use hardcoded values.

Tier 1 channels (full read):

Use a two-pass approach for each channel:

• Pass 1: slack_read_channel with oldest set to today's midnight local time (compute Unix timestamp at runtime — do not hardcode). Follow every thread with replies via slack_read_thread.
• Pass 2: slack_search_public_and_private with in:#channel-name after:YYYY-MM-DD. For any result from a prior-day thread root, call slack_read_thread for full context.

Deduplicate across passes.

Your messages (all channels including DMs):

• slack_search_public_and_private with from:<@YOUR_SLACK_ID> after:YYYY-MM-DD
• For each result, call slack_read_thread

DMs:

• slack_search_public_and_private with to:<@YOUR_SLACK_ID> after:YYYY-MM-DD
• For each high-signal person in memory/people-signal.md, also read DM history via slack_read_channel with their user ID as channel_id

Tier 2 channels (signal filter): Use slack_search_public_and_private with in:#channel-name after:YYYY-MM-DD. Include only:

• Your messages and replies
• Threads you participated in
• Decisions, announcements, deadlines, or questions directed at you
• Content topically relevant to your active projects (use Phase 1 context)

Tier 3 channels: Include only if you participated or it's a company-wide announcement.

Discard all bot/automated messages. Only include messages from real humans.

Use Phase 1 context to filter: If a meeting already surfaced an action item, don't regenerate it. Flag as "confirmed in Slack" instead.

Return: notable signals grouped by source (channel name or "DM with X"), with sender and timestamp.

---

Subagent B — Pull requests

Read memory/me.md — if GitHub CLI is not configured, skip this subagent and return nothing.

1. Run both in parallel (substitute today's date):
   • gh search prs --author "@me" --created ">=YYYY-MM-DD" --json number,title,repository,state,url,createdAt --limit 20
   • gh search prs --author "@me" --merged ">=YYYY-MM-DD" --json number,title,repository,state,url,mergedAt --limit 20
2. Deduplicate. For each PR: repo, number, title, state, one-line description.

Return: PRs created or merged today. If none, return nothing.

---

Subagent C — Project tracker updates

Read memory/projects.md for active projects and their tracker links (Asana, Linear, Jira, Notion). For each project with a tracking link, look for project-level status updates posted today or since the last working day. Do not enumerate individual tickets.

Asana: For each project with an Asana link, use get_status_overview. Only include updates from today or the previous working day.

Linear: For each project with a Linear link, use get_status_updates with type: "project" and createdAt set to the previous working day.

Jira: For each project with a Jira link, query for status changes or comments on epics/initiatives (not individual tickets) from today.

Notion: For each project with a Notion link, check for updates to the project page from today.

Skip any tracker that is not connected or returns an error — degrade gracefully.

Return: status updates grouped by project, with author, date, and 1-3 sentence summary. Flag discrepancies with Phase 1 meeting context.

---

Subagent D — Local file activity

Read memory/me.md for the workspace path.

1. Find files modified today under the workspace folder, excluding daily-notes/, meeting-notes/, and any archive directories:

   find {{WORKSPACE_PATH}} -type f -newermt "YYYY-MM-DD 00:00:00" \
     -not -path "*/daily-notes/*" \
     -not -path "*/meeting-notes/*"
   

   (Substitute today's date for YYYY-MM-DD.)
2. For each file: read in full (150-line chunks if needed). Identify: document type, project, state, 1-2 sentence summary.

Return: files modified today with path, type, project, and summary. If none, return nothing.

---

Subagent E — Custom sources (conditional)

Read memory/custom-sources.md. If the file is empty or contains no entries outside the comment block, skip this subagent entirely and return nothing.

For each entry in the file:

1. Note the source name, type, access pattern, extract instructions, filter, and relevance scope.
2. Apply the filter:
   • today — only content from today
   • since-last-working-day — today and the previous working day
   • always — no date filter
3. Execute the access pattern:
   • mcp: call the specified tool with the specified parameters
   • cli: run the command via Bash
   • local-files: glob the path, read matching files modified within the filter window
   • http: fetch the URL with the specified method and headers
4. Extract what the entry specifies. Use Phase 1 context to filter for relevance — if the entry is scoped to specific projects, only include signals that touch those projects.
5. Discard empty results silently.

Return: findings grouped by source name, with a brief summary per item. If a source errors or returns nothing, note it briefly and continue.

---

Step 1 — To-do verification pass

For every candidate to-do item, assign one of:

• KEEP — still open, no evidence of completion
• DROP — completed today or previously
• MERGE — variant of another item; combine

1. Carry-forwards: For each unchecked item, check Phase 2 signals for completion evidence. If it references a specific doc, read that file directly.

2. New items from meetings and Slack: Check whether completed same-day before including:

   • "Send X a message" → check Subagent A
   • "Update the spec" → read the file directly
   • "Open a PR" → check Subagent B
   • Decision made in the meeting itself → done, do not carry forward

3. Ownership filter: Only your own tasks. Exclude delegated work, other teams' blockers (unless they gate your work directly), same-day completions.

Output: verified to-do list with KEEP / DROP / MERGE dispositions and one-line rationale. This feeds directly into Step 2 — do not re-litigate dispositions there.

---

Step 2 — Synthesize

What happened today: 2-4 sentences on the day's 2-3 main themes. Not a list of meetings — what actually mattered and why.

Project sections: Read memory/projects.md for the active project list. Structure:

• Your team projects: always present. Write "No activity today." if quiet.
• Cross-functional projects: only if active today.
• Other: true overflow only, for items that don't fit any named project.

Each section:

1. Opening sentence: Where things stand today. The headline.
2. Top-level bullets: Decisions made, blockers surfaced, milestones reached, major open questions. High bar — if it doesn't clear this, it's a sub-bullet or dropped.
3. Sub-bullets: Supporting detail, minor flags, specifics.

Formatting:

• No em-dashes as separators. Use colons or plain sentences.
• Bold section headers (### Project Name), not bold inline labels.
• Merge signals from meetings, Slack, PRs, tracker updates, and local files into the same section if they touch the same project.

To-dos: Merge KEEP items, new AI-derived items, and manual items from today's note. Apply 5-day dedup pass against completed items. Use ownership filter before including any item.

#### Date-anchored
- [ ] [task] — [deadline or trigger] 🤖

#### Quick actions
- [ ] [task]

#### Deep work
- [ ] [task] 🤖

#### To-be-triaged
- [ ] [task]

• Date-anchored: known due date or trigger event
• Quick actions: completable in <30 min
• Deep work: requires sustained focus or collaboration
• To-be-triaged: no clear category yet — use sparingly
• 🤖: AI could meaningfully help

What I drove today: 1-3 plain bullets. High bar: decisions made, documents produced, alignment achieved, PRs shipped, problems solved. Omit section entirely if nothing clears the bar.

---

Step 3 — Write to daily note

Order:

1. To-dos block at the very top, above everything else
2. Meetings section: # Meetings with wikilinks sorted by time ascending
3. Daily Summary with project sections and "What I drove today"

Format:

### To-dos

#### Date-anchored
...

#### Quick actions
...

#### Deep work
...

#### To-be-triaged
...

# Meetings
- HH:MM - [[meeting-notes/YYYY-MM-DD/FILENAME_WITHOUT_EXTENSION|Display Name]]

## Daily Summary

### What happened today
[2-4 sentence narrative]

### [Project Name]
[One sentence: where things stand.]

- [Top-level bullet]
  - [Sub-bullet]

### What I drove today
- [plain bullet]

### Other
- [Short items that don't fit]

Handling existing content:

• Replace existing ### To-dos block entirely
• Leave existing # Meetings section as-is — do not overwrite
• Remove ## My To-dos section if present (items now in To-dos block)
• Never overwrite an existing ## Daily Summary — only append if none exists

---

Step 4 — Consistency check

1. Every meaningful signal from Phase 1 and Phase 2 appears somewhere in the note
2. Every claim traces to a source
3. Every KEEP item from Step 1 appears in the to-do list. No DROP items appear.
4. Meetings section matches Phase 1 output — every meeting file has a wikilink with correct time and display name

Fix any gaps before proceeding.

---

Step 5 — Review people and channels

People: Check today's Slack interactions against memory/people-signal.md. Add anyone you engaged with substantively 3+ times recently who isn't listed. Resolve Slack handle via slack_search_users if needed. Do not remove existing entries.

Channels: If you participated in a channel not in memory/slack-channels.md, flag it and ask whether to add it permanently.

---

Step 6 — Confirm

"Daily summary written to [path]. [N] to-dos ([M] carried forward, [K] new). [Any notable flags from consistency check or channels review.]"