Workspace focus descriptor — agent-maintained status (free, no consent)

Aperta
#70 1 commento 0 reazioni 0 assegnatari Vedi su GitHub

Nessuno ha ancora preso questa issue.

Valutazione

Difficoltà
5/5
Tempo stimato
Più di una settimana
Idoneità per principianti
45/100
Tipo di issue
Funzionalità
Chiarezza
Abbastanza chiara
Stato di attività
Tranquilla
Stack tecnologico
elixir
Ambito
full-stack

Direzione di ricerca

Inizia da plans/operator-hub.md e traccia gli entry point di overview, peek_workspace, attention view, workspace UI e workspace-agent MCP tool menzionati nell'issue. Mappa come viene memorizzato e persistito il workspace state in ETS, quindi determina ownership, comportamento degli aggiornamenti, limiti e gestione del consenso. Il lavoro è completo quando un focus descriptor delimitato viene memorizzato, reso disponibile a operatori e persone e può essere aggiornato dal workspace agent.

Scritto dal modello di indicizzazione a partire dal testo della issue.

Descrizione

feature ux

Workspace focus descriptor: agent-maintained "what we're working on" (free, no consent)

Why

Two real pains:

  1. You flip into a workspace and forget what you were doing. A short "here's
    where this is at / what we're working on" reminder, kept current by the agent
    as it progresses, fixes that.
  2. The operator (chief of staff) needs lean status per workspace without
    pulling each one's full chat — that would blow its context. A few-sentence
    descriptor per workspace is the perfect "remind me what you're working on
    again" for a busy chief of staff.

These are the same artifact: a short, living focus descriptor per workspace.

What
  • A workspace-level field (a few sentences: current focus + where we're at),
    stored on the workspace (ETS + persisted), surfaced:
    • in the workspace UI (header/sidebar) so the human is re-oriented instantly,
    • to the operator via overview / peek_workspace and the attention view —
      a cheap, lean status so the operator stays informed without reading transcripts.
  • A workspace-agent MCP tool (e.g. set_focus / update_status) the agent
    calls to update it — FREELY, no consent card (it's a description, not a
    change to the world). Think of it as the workspace agent reporting up to the
    chief of staff: "here's what I'm working on, here's where we're at."
  • The agent keeps it current — e.g. a light re-evaluation after turns ("what were
    we working on again?") and an update when the focus shifts.
Constraints (important)
  • Bounded — a few sentences, hard cap. Not a transcript, not a running log.
  • No massive token burn / context explosion — the re-eval must be cheap;
    prefer prompt guidance ("keep your focus descriptor current, briefly") over a
    forced full-summary every turn. Consider only updating when the focus actually
    changed.
  • Free to update (no consent) — distinct from rename (which is consent-gated).
Open questions
  • Is keeping it current a workspace-agent skill/prompt behavior, or something
    more structured? (Lean toward: a tool + light prompt guidance, agent-driven.)
  • Does the operator ever set it, or only read it? (Lean: workspace agent owns it;
    operator reads.)
Relates to

plans/operator-hub.md — feeds overview, peek_workspace, and the attention
view. This is the lean status backbone that keeps the operator's context small.

Lingua principale
Elixir
Stelle
85
Fork
11
Metriche di merge delle PR
Nessuna PR unita negli ultimi 30g

Preparare l'ambiente

Come iniziare

  1. Leggi tutta la issue e poi la guida ai contributi del progetto.
  2. Commenta sulla issue per dire che te ne occupi tu — evita che due persone facciano lo stesso lavoro.
  3. Fai un fork del repository e lavora su un branch.
  4. Apri una pull request che faccia riferimento al numero della issue.

Altre issue di loopyard/loopyard

Tutte le issue di loopyard/loopyard

Issue simili

Altre issue su Elixir

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.