Weekly Review Generator

Living a week and reviewing a week use different cognitive faculties. In the moment, everything feels discrete — this meeting, that article, this idea. In review, patterns emerge: the same problem surfacing in three different contexts, an idea from Monday's reading that would have solved Thursday's challenge. The reviewer-from-inside almost never sees these. You're too close. The skill earns its keep by being the bird's-eye view you can't generate on your own.

Five lenses run sequentially. Theme extraction — recurring topics, concerns, or concepts across the week, separating surface themes ("I had a lot of meetings about the launch") from deep themes ("three separate conversations pointed to the same organisational trust deficit and nobody has connected them"). Connection discovery — ideas from different contexts that relate but weren't linked in real time. This is the highest-value lens, and it's the one no human can run on their own notes effectively. Evolution tracking — how thinking shifted across the week, which positions changed, which questions got sharper. Energy and emotional patterns — only applied when reflective content is actually present; the skill doesn't ask probing personal questions about a notes file that's all tasks. Forward threads — specific, named questions to carry into next week.

The skill adapts to what's actually there. A light week (one or two notes, thin content) gets themes and forward threads only, framed honestly — "not enough material for a full review, here's what stands out." A task-only week gets activity review mode plus a nudge: "I can see what you did but not what you thought about it; one sentence of reflection per day would change what next week's review can do." A dense week gets the full five lenses. The skill refuses to invent insight where the input has none — fabricating a "deep theme" from three operational meetings is exactly the kind of slop that makes people quit weekly reviews.

When the user shares last week's review, open threads get tracked explicitly. Each one is either advanced ("you encountered two relevant data points this week"), unchanged ("no new evidence, still worth watching"), or — if it's been carrying for three weeks — flagged as systemic ("this isn't a one-off, it's a standing concern"). The output is narrative, not bullet-soup. A list of things you noted is a table of contents; a story that weaves them into a coherent week is a review. The skill tells the story.