mirror of
https://github.com/pocketpaw/pocketpaw.git
synced 2026-05-18 16:06:22 +00:00
c70d8e45b3
Two companion docs to the existing mission-control spec, written off real ground truth from ee/instinct/ and ee/fabric/: backend-audit.md — primitive-by-primitive accounting of what exists vs what needs to be built. Honest finding: ~60% of Mission Control wires to endpoints that already ship. The Tray + Pawprints + per-item audit are essentially there via ee/instinct/router.py (propose / pending_actions / approve / reject / query_audit / get_audit_entry). Fabric, Pockets, Soul, Connectors, Notifications all in place. The two genuinely new entities: Tasks (unified WorkItem with agent + human assignee polymorphism, cycle assignment, claim/complete flow) and Cycles (time-boxed windows with daily snapshot job for the burnup chart). Plus one façade entity (ee/cloud/mission_control/) for the single-round-trip read pattern, one activity buffer for the live ticker, and a bulk-approve endpoint on instinct. Three PRs sized roughly 12-18 agent-hours of crew work. user-flows.md — seven concrete end-to-end journeys through Mission Control, grounded in Shawn (events-production CTO, 25 staff, $6-10M ARR). Monday-morning sweep, mid-day Nudge approval, delegating to an agent (the moneyshot for the agentic pitch), bulk approve, multi- coordinator handoff to Jess, complaint investigation via Pawprints (the C-suite-buyer-closer demo), Friday cycle + analytics review. For each: trigger, screen state, click path, primitives touched, backend round-trips, success state, failure modes to guard against. Plus cross-cutting interactions (assistant sidebar behavior, keyboard shortcuts, notification fan-out policy) and an explicit list of what we deliberately don't build in v0.5/v1.