Phase 3, Triage suggestions: every few minutes the work service's cron asks the workspace's own AI provider (integrations' ai_brief, on its fast model; g1t runs no model of its own) where each new work item in Triage goes: a horizon (now, next or later), an appetite, an open work item it seems to repeat, and why. Asked once per work item, a dozen calls a run at most; a workspace with no provider gets none and holds no one else up. Triage shows the suggestion with Accept and, for a likely duplicate, Close as duplicate (a "Duplicate of #N" comment, then closed as not planned); an item's Plan box shows it too. Home tells those who can plan when work items have waited in Triage more than a week, after the work to take next.
Phase 4, progress draws itself: get_plan answers progress for each work item in Now: its hill (figuring_out, decided, building, landing) from its decisions, commits and pull requests, when anything last happened, quiet after 5 days, and a sentence or two on where it stands that g1t drafts from its sessions, decisions and pull requests whenever something new happens. The Plan's Now rows and an item's Plan box show them; This week lists what landed from Now in the last 7 days and how many went quiet. g1t work plan (CLI 0.4.3) prints the hill, the standing and what landed.
Phase 5, letting go: a work item in Later with nothing happening for 90 days is set aside on the cron, saying so in its conversation and publishing work_item.planned with no actor; the plan says how many were set aside in the last 30 days.
Work service migration 0057_plan_reads_itself (work_items.suggestion, suggested_at, standing, standing_at). Docs: work/plan (suggestions, how work in Now stands, letting go), guides/ai (what the plan asks the provider), guides/home, the get_plan reference.
Checked: cargo test -p g1t-work --lib (302, new: the hill from the record, quiet, what waits too long, reading a suggestion, bounded standings), -p g1t-contracts --lib, -p g1t-api (222), -p g1t (117 unit); cargo check --workspace --tests; cargo check -p g1t-api -p g1t-work --target wasm32-unknown-unknown; apps/web npm run typecheck, npm test (1121), react-router build, load-check (every route loads, 281); Prettier; docs build. Pages rendered with mock data at desktop and 375px, dark and light. G1t-Session: ses_01m5bgsckce7gt943v64h219yy