Week Plan

<!-- Generated from `.claude/skills/week-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.

Purpose

Set priorities and plan the week ahead. Now with intelligent priority suggestions based on quarterly goals, calendar capacity, and task effort classification.

Usage

• /week-plan — Plan current week (or next week if run on the last working day or a non-working day)
• /week-plan next — Explicitly plan next week
• /week-plan current — Force planning current week

When to Use

Best times:

• First working day, before work - Before diving into daily work
• Last working day, after work - Set up next week while context is fresh
• Evening before the first working day - Plan before the week begins

Read working_week.days in System/user-profile.yaml before deciding which days are first, last, or outside the working week.

Step 1: Determine Target Week

Calculate target week (current or next) based on day of week and user parameter.

Step 2: Context Gathering (ENHANCED)

Gather comprehensive context to inform intelligent priority suggestions.

2.1 Last Week's Review

Check for 00-Inbox/Weekly_Synthesis_[last-monday].md:

• "Next Week" section → Suggested priorities
• "Carried Over" section → Unfinished tasks
• "Blocked Items" → Things that need resolution
• "Learnings" → Insights to apply

2.2 Quarterly Goals Status

Use: get_quarterly_goals()
Use: get_goal_status(goal_id) for each goal that has a goal_id

Skip get_goal_status for any goal whose goal_id is null — it has no ID to look up, and the call will tell you so rather than return a status.

For each goal, get:

• Current progress (concrete: "2 of 5 milestones complete")
• Linked priorities count
• Weeks since last activity
• Stall warnings

Check activity_known before judging any goal. When it is false — the goal was typed by hand with no ID, or recovered from a freeform list — linked_priorities and linked_priorities_count are null, meaning not known, never none. Never call such a goal orphaned or stalled on that basis.

Identify goals needing attention (only among goals where activity_known is true):

• Goals with no linked priorities (orphaned)
• Goals with no activity in 2+ weeks (stalled)
• Goals behind expected pace

If the response flags goals as provisional (recovered from a freeform list), offer once — "Want me to structure these into trackable goals?" — and move on without nagging if declined.

2.3 Open Tasks and Projects

Use: list_tasks(include_done=False)

Get all open tasks and:

• Classify by effort (deep_work / medium / quick)
• Group by pillar alignment
• Identify P0/P1 items

Goal-advancing tasks — read the groomed backlog, don't guess:

Use: get_weekly_planning_context()

Each goal in the response's goal_health carries open_task_count and next_up_tasks — the top items the user already marked "take these first" during grooming.

First, check activity_known. When it is false, that goal has no usable ID — it was typed by hand, or recovered from a freeform list — so Dex cannot tell what links to it. linked_priority_count, has_activity, open_task_count and next_up_tasks are all null for that goal, meaning not known, never none. Do not read a null as zero, do not call it stalled or neglected, and do not call get_goal_backlog for it (a null ID falls back to every goal's backlog). Plan around it normally and pass on the recommendations line that says how to give it an ID.

For every goal where activity_known is true, use the fields directly:

• Goal has next_up_tasks: suggest those items by name — "Goal X's next-up items are A and B — they fit Wednesday's block."
• Stalled goal with open tasks but no groomed order: call get_goal_backlog(goal_id) and pick candidates from the top of its ordering (priority, then oldest).
• Goal with a large open_task_count (10+) and no next_up_tasks: mention /goal-backlog once — "Goal X has 14 open tasks and no groomed order — worth a /goal-backlog pass after planning?" Say it once for the whole plan, not per goal, and drop it if declined.

2.4 Calendar Shape Analysis (NEW)

Read calendar.provider from System/user-profile.yaml before any calendar call.

If provider is google: use Google Workspace MCP, not calendar-mcp event tools.

Read calendar.work_calendar from the same profile. Call list_calendars(detailed=true) first. Match work_calendar against a returned calendar summary, id, or primary flag, then pass that calendar's id. If work_calendar is missing or primary, use primary. Do not pass a display name as calendar_id. Set max_results high enough that a stacked week is not truncated.

Use: list_calendars(detailed=true)
Use: get_events(time_min="[week start]", time_max="[day after week end]", calendar_id="<resolved id or primary>", max_results=2500, detailed=true)
Use: analyze_calendar_capacity(days_ahead=7, events=[...mapped from get_events...])

Map returned events into the shape analyze_calendar_capacity expects (title, date as YYYY-MM-DD, duration_minutes, attendees, all_day). Apply CLAUDE.md's Calendar response confidence contract to the outcome: a missing or failing Google Workspace calendar tool is not an empty week, and do not call analyze_calendar_capacity with missing results. If the tools are unavailable or not connected, point to /google-workspace-setup using that skill's existing copy. Do not invent new privacy or consent language.

If provider is none: skip calendar gathering. Do not call calendar-mcp event tools. Do not treat the week as open.

If provider is apple, missing, or any other value: use calendar-mcp:

Use: calendar_get_events_with_attendees(start_date="[week start]", end_date="[day after week end]")
Use: analyze_calendar_capacity(days_ahead=7, events=[...from calendar MCP...])

Apply CLAUDE.md's Calendar response confidence contract before consuming events or passing them to analyze_calendar_capacity.

Understand the shape of the week:

Week capacity summary:

• Deep work opportunities: {{count}} days with 2+ hour blocks
• Total deep work hours available: ~{{X}} hours
• Stacked days to avoid scheduling deep work: {{list}}

2.5 Task-to-Time Matching (NEW)

Use: classify_task_effort(title) for key tasks
Use: suggest_task_scheduling(include_all_tasks=True, calendar_events=[...])

Get intelligent matching:

• Which tasks need deep work time?
• Which can fit in gaps?
• Are there enough slots for all deep work?

Capacity check:

"You have 5 deep work tasks totaling ~12 hours. You have 2 open days with ~6 hours of deep work capacity.

⚠️ Capacity gap: Consider deferring 2 deep work items or protecting more time."