Daily Plan

<!-- Generated from `.claude/skills/daily-plan/SKILL.md` by `scripts/generate-agents-skills.py`. Do not edit. -->

Execution mode

Run inline in the current conversation by default, so this work can see what the user has already discussed, decided, or settled this session. Do not fork merely because this skill was selected. Only run in the background when the user explicitly asks for a background run or the host has already obtained a specific background-work approval for this run.

Delegated gathering (large-vault scaling)

This skill stays inline as described above: it keeps session awareness, it asks the user the questions, and it owns every interactive step. What it does NOT do inline is the bulk read-gathering, which on a mature vault (hundreds of notes, thousands of indexed messages, a live calendar and multiple integrations) can be large enough to exhaust the main conversation before the useful work starts.

So the gathering phase is delegated to one general-purpose subagent via the Agent tool, using the self-contained prompt in this skill's AGENT_INSTRUCTIONS.md:

1. Read .agents/skills/daily-plan/AGENT_INSTRUCTIONS.md.
2. Substitute its placeholders ({{TARGET_DATE}}, {{TARGET_DATE_PLUS_1}}, {{DAY_NAME}}, {{MONTH}}, {{DD}}).
3. Call the Agent tool with subagent_type: "general-purpose", model: "sonnet", that prompt, and a short description. The brief is mechanical gathering, so it runs on the fast tier declared by this skill's model_routing.steps.data-gathering; the judgement calls stay in this conversation on the default model.
4. Verify it wrote the draft plan to the resolved daily-plan folder (see Daily plan folder), then run the remaining interactive steps from its findings and present the plan.
5. Close out every <!-- NEEDS TASK --> line in the draft. The subagent never creates tasks, so a focus candidate with no existing task is written without a checkbox and marked. For each one: confirm it with the user per Step 6, create the task with create_task, then rewrite that line as a - [ ] checkbox carrying the returned ^task-YYYYMMDD-XXX anchor and remove the marker. If the user declines, remove the marker and leave the line checkbox-free. Never leave the marker in the saved plan, and never turn one of these lines into a checkbox without an anchor — an anchorless checkbox is invisible to completion sync forever (Step 6).

The subagent inherits MCP connections, runs in its own context, and that context is freed when it completes, so only its findings reach this conversation.

Use AGENT_INSTRUCTIONS.md verbatim. Read the file and pass its content as the subagent prompt, substituting only the placeholders. Do NOT hand-write a replacement brief from what you already know about the day: that is how steps get silently dropped, and the omission looks complete because nothing errors. If context from this conversation is worth adding, APPEND it to the file's content; never substitute for it.

Two caveats that are load-bearing:

• Do not count on hooks for the subagent's writes. The hooks declared in this skill's own frontmatter belong to this skill's run, not the subagent's, and whether the repository-wide hooks in the host's lifecycle settings reach a subagent's tool calls is not something a skill should assume either way. Nothing in this skill's gathering depends on a hook; the subagent's writes must stand on their own.
• Always fall back. If the subagent fails, times out, or returns nothing usable, say so plainly and run the gathering inline from the same AGENT_INSTRUCTIONS.md. A missing subagent must never mean a missing result.

Stays inline: the journal and Monday week-planning gates, the yesterday's review check, Dex Inbox triage, the meeting-task review and goal-link confirmations, inbound external task review, pushing focus tasks to Reminders, and usage tracking. These need the user, so they must not be delegated.

Previous-working-day notes (Step 5.10c) are gathering, not the yesterday's review check. That step lives in AGENT_INSTRUCTIONS.md so a customization of it reaches the subagent. The written plan must show the post-condition: notes found for a named date/file, or none.

Purpose

Generate your daily plan with full context awareness. Automatically gathers information from your calendar, tasks, meetings, relationships, and weekly progress to create a focused plan with genuine situational awareness.

Usage

• /daily-plan — Create today's daily plan
• /daily-plan tomorrow — Plan for tomorrow (evening planning)
• /daily-plan --setup — Re-run integration setup

Tone Calibration

Before executing this command, read System/user-profile.yaml → communication section and adapt tone accordingly (see CLAUDE.md → "Communication Adaptation").

Step 0: Process Unprocessed Meetings (only when something is waiting)

Before gathering context, ask the session-start sweep how many meeting records are still waiting, without re-scanning the vault yourself. The count mode prints one integer, bypasses the sweep's once-per-30-minutes notice throttle, and writes nothing, so it is safe to call every morning:

Ask the host for its count of waiting meeting records if it keeps one; if it does not, treat the count as unknown and run the meeting pass.

Branch on the answer:

• 0: skip this step entirely and say nothing. Yesterday evening's review already ran the catch-up pass, so on most mornings this is the answer, and re-running a full seven-day /process-meetings here found nothing new while costing minutes.
• Above 0: run /process-meetings. This creates meeting notes, updates person/company pages, and extracts tasks, so the plan has complete data. Note the count and use the extracted context in the plan. Do NOT ask for a skill rating after this sub-step — save that for the end of the full plan.
• The command fails or prints nothing: treat that as unknown, not as zero: run /process-meetings as before.

Step 0.5: Dex Inbox Check (Phone Captures)

Check for tasks captured from phone that haven't been triaged:

Use: reminders_list_items(list_name="Dex Inbox")

If the tool is unavailable or errors (Apple Reminders phone-capture is optional and may not be set up on this machine): skip this step silently — do not surface an error for a feature the user never enabled.

If items found, triage them before building the plan so task counts are accurate. See Step 5.10a for the full triage flow.

If empty: Skip silently.

Step 1: Background Checks (Silent)

Run these silently without user-facing output:

1. Release evidence: SessionStart already performs the at-most-daily bounded fetch-only check. Read its result; do not start a second network attempt. Surface only its complete release-appears-available-unverified notice, verbatim. Keep no-newer-release-observed-unverified, offline, UNKNOWN, and skipped silent here.
2. Self-learning checks: Run changelog and learning review scripts if due
3. Search index refresh: Run qmd update && qmd embed to refresh vault search index with any overnight changes (meetings processed, files edited, etc.). If qmd is not installed, skip silently.
4. People index refresh: Call build_people_index from Work MCP. This keeps the People Directory current so person lookups throughout the day are fast. Takes <2 seconds.
5. Innovation synthesis (silent): Call synthesize_changelog() and synthesize_learnings() from Improvements MCP. These run in background and populate the backlog — results are surfaced in Step 1.5 below.
6. Granola check (silent): Granola sync uses the official API via GRANOLA_API_KEY (environment or vault-root .env). If no key is configured, skip all Granola steps silently — an unconnected optional integration is not an error.
7. System health (silent): Read System/.smoke-last-run.json when present and retain any broken-journey context for Step 5.10b.

Step 1.5: Innovation Spotlight (Concierge)

After background checks complete, check for noteworthy backlog activity:

1. Call list_ideas(status="active", min_score=70) from Improvements MCP
2. Check System/.synthesis-state.json for recent synthesis activity (last 7 days)
3. If there are AI-authored or recently enriched ideas, pick the most impactful one

Surface as a brief spotlight in the plan output (1-2 lines max):

Innovation Spotlight: Claude Code shipped native memory (v2.1.32) — this could simplify idea-006 (Session Memory MCP). Run /dex-improve idea-006 to explore.

Rules:

• Show at most 1 spotlight per daily plan (don't overwhelm)
• Rotate through ideas — don't show the same one twice in a row
• Only show if there's genuine "Why Now?" urgency (new evidence in last 7 days)
• If no recent synthesis activity, skip this section entirely
• Never block the plan for this — it's a helpful aside, not a gate

Step 2: Morning Journal Check (If Enabled)

If journaling.morning: true in user-profile.yaml, check for today's morning journal and prompt if missing.

Step 3: Monday Weekly Planning Gate

If today is Monday and week isn't planned, offer to run /week-plan first.

Step 4: Yesterday's Review Check (Soft Gate)

Check for yesterday's review and extract context (open loops, tomorrow's focus, blocked items).

Step 5: Context Gathering (ENHANCED)

Gather context from all available sources. This is where the magic happens.

5.1 Midweek Progress Check (NEW)

Use: get_week_progress()

This is critical for genuine situational awareness. Extract:

• Day of week and days remaining
• Weekly priority status (complete / in_progress / not_started)
• Warnings for priorities with no activity

Surface this prominently:

"It's Wednesday. Here's where you are on this week's priorities:

1. ✅ Ship pricing page — Complete (finished Monday)
2. 🔄 Review proposal — In progress (2 of 5 tasks done)
3. ⚠️ Customer interviews — Not started (no activity yet)

You have 2 days left this week. Priority 3 needs attention."

5.2 Calendar Capacity Analysis (NEW)

Calendar events come from the delegated gatherer. That gatherer reads calendar.provider in System/user-profile.yaml and uses Google Workspace MCP when the provider is google, or calendar-mcp when it is not.

Use: analyze_calendar_capacity(days_ahead=1, events=[...from the configured calendar source...])

Understand the shape of today:

• Day type: stacked / moderate / open
• Meeting count and hours
• Free blocks available
• Recommendation: What kind of work fits today

Surface this:

"📅 Today's shape: Moderate (4 meetings, 3 hours total)

Free blocks:

• 8:00-9:30 AM (90 min) — Morning focus time
• 2:00-4:00 PM (120 min) — Afternoon block

Recommendation: Good for medium tasks and meeting prep. Deep work fits the 2-4pm block."

5.3 Meeting Intelligence (NEW)

For each meeting today:

Use: get_meeting_context(meeting_title="...", attendees=[...])

Get genuine context, not just attendee names:

• Related project: What project is this connected to?
• Project status: What's outstanding? What's blocked?
• Outstanding tasks with attendees: What do you owe them? What do they owe you?
• Prep suggestions: What should you review before this meeting?

Surface this with surprise and delight:

"📍 Meeting: Acme Quarterly Review (2pm with Sarah Chen, Mike Ross)

Related project: Acme Implementation (Phase 2)

• Status: On track, but pricing section still in draft
• Outstanding: You owe Sarah the pricing proposal

Prep suggestion: Review proposal draft, prepare pricing options. Block 30 min before this meeting?"

5.4 Commitment Tracking (NEW)

Before commitments, read System/.dex/entity-suggestions.json (if present). If it has suggested entries, add one compact context line: "Dex suggests person pages for: X (N meetings), Y — say the word and I'll create them." This step is read-only.

Also read System/.dex/entity-cooling.json when it is present and fresh. If its consequential cold list is non-empty, surface it as one compact "❄️ Going cold" line in ## ⚠️ Heads Up; degrade silently when the feed is absent or empty, and never widen this into a vault-wide people dump.

Also read System/.dex/entity-relationships.json when it is present and fresh. If its suggestions list is non-empty, surface one compact "🔗 Relationships to confirm" line in ## ⚠️ Heads Up; degrade silently when the feed is absent, stale, or empty. Keep it as a nudge, show the page and edge being proposed, and offer the real Work MCP actions: confirm_relationship(page, edge_key) or dismiss_relationship(page, edge_key). Never choose either action without the user's per-edge say-so.

Use: get_commitments_due(date_range="today")

Surface things you said you'd do:

"⚡ Commitments due today:

• You told Mike you'd get back to him by Wednesday (from Monday 1:1)
• Follow up on competitive analysis (from Acme meeting)"

5.5 Task Scheduling Suggestions (NEW)

Use: suggest_task_scheduling(include_all_tasks=False, calendar_events=[...])

Match tasks to available time based on effort classification:

"📋 Scheduling suggestions:

Task	Effort	Suggested Time
Write Q1 strategy doc	Deep work (2-3h)	Tomorrow (you have a 3h morning block)
Review Sarah's proposal	Medium (1h)	Today 2-3pm (before Acme meeting)
Reply to Mike	Quick (15min)	Between meetings

⚠️ Heads up: You have 2 deep work tasks but today's too fragmented. Consider protecting tomorrow morning."

5.6 Semantic Context Enrichment (if QMD available)

This step runs automatically when QMD is installed via /enable-semantic-search. It adds a semantic search layer on top of the standard context gathering to surface connections that keyword search misses.

Check if QMD MCP tools are available by calling the status tool (QMD MCP). If available:

1. For each meeting today, run:

   query(query="[meeting topic] [attendee names]", limit=3)

   Surface: past discussions, related decisions, relevant commitments that share meaning but not keywords with this meeting. Example: a meeting about "customer onboarding" finds notes about "activation rates" and "time to value".

2. For each weekly priority that's lagging, run:

   query(query="[priority description]", limit=3)

   Surface: vault content that advances or relates to this priority but wouldn't appear in a keyword search. Especially useful for finding forgotten context about stalled work.

3. Cross-topic connection scan:

   query(query="[today's key themes combined]", limit=5)

   Surface: unexpected connections between today's meetings, tasks, and priorities. This is where semantic search shines — finding that a 2pm customer call relates to a PRD you wrote last month using completely different terminology.

4. Merge with existing context — only add genuinely new insights. Don't duplicate what Steps 5.1-5.5 already found. Mark semantic results with their source so the plan output can distinguish them.

What this enables in the plan output:

• Meeting context sections include "Also relevant:" with thematically related past discussions
• Priority recommendations cite relevant vault content discovered by meaning
• "Heads Up" section catches connections between seemingly unrelated items
• Focus recommendations are informed by deeper vault knowledge

If QMD is not available: Skip silently. Steps 5.1-5.5 and 5.7 provide full context via standard methods.

5.7 Reminders Completion Sync (Dex Today → Dex)

Check if any tasks were completed on phone since the last plan:

Use: reminders_list_completed(list_name="Dex Today")

If the tool is unavailable or errors (Apple Reminders sync is optional and may not be set up on this machine): skip this step silently — do not surface an error for a feature the user never enabled. Note: Reminders access never works when Claude Code runs inside the VS Code extension (macOS never shows the permission dialog to that process) — see the known limitation in 06-Resources/Dex_System/Calendar_Setup.md. Do not advise reinstalling or reconfiguring; skip silently.

For each completed item:

• Match to a Dex task by title
• Update task status via Work MCP: update_task_status(task_title="...", status="d")
• Surface what was synced:

"📱 Synced from phone:

• ✅ "Follow up with Hero Coders" — marked done in Dex"

If nothing to sync: Skip silently.

5.8 Email Intelligence (if connected)

Check System/integrations/config.yaml for google-workspace.enabled: true. Also treat a registered apple-mail-mcp server as a connected source. Before querying Apple Mail, run python3 core/utils/doctor.py --deep --only mail.apple-search; Apple Mail search is usable only when the mail.apple-search check reports OK / feature_status: ok. --only runs that one probe instead of the whole live checkup (smoke journeys, search index, connected tools), which is what a full --deep run does and what made this step slow every morning. Google Workspace needs no local probe: its own tool response carries the health status.

If connected and healthy:

1. Use an attention-inbox count for the headline, never a provider-wide unread total. For Google Workspace, count unread in the attention inbox. Start with is:unread category:primary. If that returns 0, confirm with is:unread in:inbox and whether any mail has the Primary label (category:primary, not only unread). When inbox tabs are off, category:primary is empty even though the inbox has unread mail — use the inbox count. When Primary exists but has no unread, report 0. Never use a provider-wide is:unread total. Use monitored labels separately to find priority messages. For Apple Mail, count unread messages in Inbox mailboxes. If the source cannot scope this reliably, omit the headline unread count.
2. Flag emails needing reply (> 48h since received, from key contacts in 05-Areas/People/)
3. Surface email threads with today's meeting attendees

Include in plan:

"Email: [X] unread in Primary/Inbox, [Y] need replies. [Z] threads with today's meeting attendees."

For Apple Mail, never interpret an empty search as "no matching mail" unless that health check is OK. If a connected source is broken or could not be checked, do not silently skip: include one calm "Email context omitted" line with Doctor's user_message or fix path. If the source is not connected (OFF), omit it without noise.

5.9 Teams Intelligence (if Teams connected)

Check System/integrations/config.yaml for teams.enabled: true.

If enabled and MCP healthy:

1. Get unread messages from priority channels
2. Surface DMs needing response
3. Check for mentions

Include in plan:

"Teams: [X] unread chats, [Y] mentions. [Z] threads with today's meeting attendees."

If BOTH Slack and Teams enabled:

• Show both digests, clearly labeled: "Slack: ..." and "Teams: ..."
• Deduplicate if the same person appears in both (merge context, label the source)
• Present side by side in the plan output under a combined "Chat Intelligence" heading

If unhealthy: skip silently (graceful degradation -- no error to user).