Story: Agile timeline: bucketed summaries of recent activity

With several environments and LLM sessions working in parallel, it is hard to keep up with what happened and to catch problems as they arise. This story delivers a timeline: a compass timeline command that lists the documents changed in a period (e.g. "timeline now" → last 20 minutes); an LLM skill that buckets those changes and writes a snapshot per time window — "environment local1 worked on story X, tasks Y1, Y2; this is a new story not in the sprint; task Y1 had problem P0" — highlighting problems and suspicious LLM decisions; and a timeline view in the agile board showing event counts over time with click-through to each bucket's log.

• compass timeline has two behaviours: generate — list stories/tasks/captures changed within a given window (with a "now" convenience for the recent past), what changed, and links: the input for the LLM snapshot; and view — show the last N buckets sourced purely from the sprint's timeline/ folder.
• Documents record which environment touched them, so snapshots can attribute work per environment.
• Snapshots are org files under the sprint's timeline/ folder, named <ISO-from>-<ISO-to>.org, bucketed (stories raised, tasks raised, worked on, problems, suspicious decisions) — tables over prose where possible, with id links to the documents.
• The agile board has a timeline view: counts of events per time bucket as a clickable chart; clicking a bucket shows its snapshot; defaults to the latest bucket.

One temporal model, two substrates

See the temporal command coherence investigation: temporal commands sit on a 2×2 grid (subject: me/everyone × time: now/past); timeline is the missing "everyone × past" quadrant. Now-views (fleet, journal) read the live substrate (worktree journals, branch/PR state); the timeline reads consistent data only — the agile docs on origin/main, their git history, and the gh PR record — so any environment reproduces the same timeline from a fresh checkout. Journals are never a timeline input; environment attribution comes from the document stamps; every snapshot carries a window-scoped audit section reporting bookkeeping drift.

Board timeline view needs no infrastructure extension

Verified: snapshot files carry :ID:=s, so =org-roam-db-sync indexes them and the existing graphdata.json export delivers their full org content to the browser — the board view consumes them exactly like stories and tasks. No changes to the export, the site build, or the data pipeline.

Bucket size is an experiment, not a constant

20 minutes was an arbitrary starting point. The real constraint is readability: a bucket must not hold more events than can be scanned at a glance. Window size is configurable in both the command and the skill; tune it with real data, and consider event-count-capped buckets (split a window when it exceeds N events) over fixed durations if busy periods overwhelm.

• Real-time push/streaming updates — snapshots are written on demand or on a cadence, not live.
• Timeline for non-agile documents (knowledge, manuals) — agile docs first; generalise later if useful.
• Alerting/notifications — the timeline is a pull-based review tool.