Local attention queue

synapse attention projects approval requests, failed delivery and recovery from a hub event store. Optionally, it reads the owner-local account and quota ledger for stale observations and exhausted windows with a recorded renewal. It can also project owner-local review feedback for missing author bindings, pending independent decisions and route receipts. The queue is advisory: it never decides an approval, retries a delivery or spends a quota. A late observer is shown as missing_observer; an empty queue is quiet only after a recent successful observation.

Use an existing hub database. A sync reads only the event kinds needed for attention, folds each reason and subject to its latest source revision, then reconciles the owner-only SQLite queue. The default queue path is $XDG_STATE_HOME/synapse-channel/attention/queue.sqlite3, or ~/.local/state/synapse-channel/attention/queue.sqlite3 when that variable is unset. Its directory and database file must be owner-only.

synapse attention sync /path/to/hub.db
synapse attention sync /path/to/hub.db --entitlement-store /path/to/ledger.sqlite3
synapse attention sync /path/to/hub.db --review-store /private/reviews.sqlite3 \
  --reviewer-seat REVIEWER/seat
synapse attention list
synapse attention list --json
synapse attention snooze approval:TASK-7 --hours 2
synapse attention resolve approval:TASK-7

--db-key-file reads an encrypted hub through the same SQLCipher key material as other local event-store commands. --approval-hours sets the review deadline (24 hours by default); when it passes, the queue marks an undecided request expired and raises its severity. Elapsed time never grants or denies approval. Use synapse approval decide to record the actual decision. An explicit local resolution only clears the alert at its current source revision. A fresh request or other new evidence reopens it. A deferred delivery acknowledgement replaces the failed-delivery alert with one time-bounded recovery item. Snoozes hide the alert until their deadline; source changes cancel the snooze. An absent C04 observation is stale, while a fresh observation clears that stale alert. --stale-hours sets its threshold.

Run sync regularly to keep the observer current. list reports the last successful source observations and defaults to a five-minute freshness limit; --observer-seconds changes that limit. If no sync has run, or one stops, the CLI and cockpit report a missing observer. A failed source read does not replace the last successful observation with a false quiet state.

Desktop delivery is opt-in. Add --desktop to sync on a host with notify-send; --desktop-max bounds each pass to at most 50 new alert revisions (default 10). One generic preview says only how many items need review. It contains no task body, account label, subject or source path. Repeated syncs suppress the same revision and severity state. A failed desktop command is not marked delivered, so a later pass can retry.

To show this same local queue in the authenticated cockpit, start the dashboard with --attention-store /path/to/queue.sqlite3. Its bearer-gated /attention.json route serves a bounded, content-minimised view. The cockpit keeps its live fleet signals and adds these persisted alerts to the same attention panel. It marks a missing or unavailable observer explicitly. The cockpit remains a read-side display; use the CLI for snooze and resolution. The dashboard does not start a sync process or read the private entitlement ledger on its own. Mobile and chat delivery are separate adapters.