Skill: weekly-review

Goal: give an honest read on the week that's ending — real vs. committed, vs. objectives, vs. learnings, vs. last week — and hand back a pre-filled seed so filling next week's plan is fast. This is the retro/dashboard that can feed a weekly planner's Step 0.

Read-only. Never creates, edits, colors or deletes calendar events. It only reads and writes markdown.

Personalize by pointing the file references below at your own docs. A private, customized copy of this skill (with your real context) is what you install and consume; keep this generic template shareable.

Inputs (read, don't dump)

1. What actually happened — not just the calendar. Cross three sources so the retro reflects the live week, not the stale Friday-of-last-week snapshot:
   • Calendar of the week ending: what got scheduled and (where visible) what moved or dropped.
   • WEEKLY_PLAN.md → "actual vs. planned" section — the live trace your daily operator left during the week (items that came in, things that got bumped, reprioritizations). This captures decisions the calendar doesn't show. If it's empty, say so — that's a signal fine-grained tracking was lost.
   • Learnings added mid-week — new entries in LEARNINGS.md logged while operating day to day, not only last Friday's distillation.
2. What was committed — WEEKLY_PLAN.md (the approved plan + energy throttle): P0/P1, one-outcome and experiment as approved. Source of truth is the markdown, not any seed note. Comparing 1 (what happened) against 2 (what was committed) is the core of the verdict.
3. Objectives — your goals doc (GOALS.md): did the week move the north star or just stay busy?
4. Learnings — LEARNINGS.md: which past pattern repeated or broke this week? (The top includes what was logged during the week, see input 1.)
5. Last week — the prior entry: compare energy, one-outcome and load.

Output — two parts

Part A — The verdict (dashboard)

Short and frank. No empty optimism. Four cuts:

• Real vs. committed — of the P0/P1 committed, what got done, what slipped, and why (energy, interruptions, over-scoping).
• Vs. objectives — did the week advance the goals, or was it maintenance?
• Vs. learnings — which pattern from LEARNINGS.md showed up again (or was finally broken).
• Vs. last week — energy, load and one-outcome vs. last week (better / same / worse, one line each). Lead with the single most important signal. Flag 🚩 what's at risk if it continues.

Part B — Seed for next week (speeds up planning)

Pre-fill the inputs you'd otherwise type from scratch, marked as editable suggestions:

• Suggested energy throttle (1–5) — based on how the last 1–2 weeks actually went.
• Carryover P0/P1 — what didn't get done and should roll forward (note if it should be dropped instead).
• One-outcome candidates — 1–2 "must-win" items tied to purpose/mastery, not just urgency.
• Watch-outs — any guardrail or eval that failed this week and must be protected next week.

Write Part B in the shape of WEEKLY_PLAN.md so the planner (or you) can drop it straight in.

Rules

• Don't invent. If the calendar or plan doesn't show something, say "no data", don't guess.
• Honest, not flattering. Name what didn't work and why. No forced productivity.
• Read-only. Never touch the calendar. Markdown only.
• Feeds the planner (gate). Save the review as a dated file with a status line (e.g. Status: PENDING VALIDATION). A weekly planner's Step 0 should not build the plan until you validate the review (flip the status to VALIDATED). The concept: validate the retro first, then plan.

How it runs

• On demand: /weekly-review.
• Scheduled (optional): as an automatic task before your weekly planning run.