Store Pulse

Store Pulse

Daily-driver ecommerce snapshot. Every invocation leads with a fresh inline answer (24h
summary, KPI row, funnel). Saving a dashboard artifact and scheduling a recurring report
are explicit, optional actions — never prerequisites.

One dashboard + one schedule per domain. Different domains can each have their own,
side by side.

Default flow

Runs on every invocation ("how's my store doing", "check on flyinmiata.com", a bare
/store-pulse). No existence check gates this and no list_artifacts call here —
always fresh, and never mentions an existing saved dashboard even if one exists (see
"Save as artifact" for the one place existence is checked).

  1. Resolve the domain — use it without asking if already clear from context;
    otherwise noibu_list_domains (one result → use it; multiple with none named →
    AskUserQuestion).
  2. Read references/inline-snapshot.md and follow it exactly — it owns the full
    response shape (summary, widget, closing line, investigate-thread question).

Save as artifact

Triggered by the snapshot widget's "Save as artifact" button (arrives as
Save Store Pulse dashboard for [domain]) or a direct request ("save this as a
dashboard").

  1. Check for an existing dashboard. list_artifacts — match on id/title containing
    store-pulse-[domain-slug], or a legacy store-pulse-dashboard whose
    SP_CONFIG.domain.name matches (see Migration note; leave its id alone).

    • Same call also decides the title: drop the domain suffix by default
      (Store Pulse — Atari), add it back only on a Label collision with another live
      dashboard (e.g. atari.ca already exists) — see references/live-dashboard.md.
  2. Exists? Ask: "You already have one, do you want to edit it?"

    • Yes → ask what to change, go to "Edit an existing dashboard".
    • No → acknowledge, stop.
  3. Doesn't exist? Ask once about optional blocks (multi-select, all default-off;
    core-kpis/purchase-funnel are always included, don't ask about those):

    "Your dashboard will include Core KPIs (sessions, engagement, conversion, AOV,
    revenue per session) and a Purchase Funnel by default. Anything else you'd like
    to add?"

    • Top products — top 5 by traffic with add-to-cart rate
    • Channel performance — sessions and conversion by traffic source
    • Paid ad performance — sessions, conversions, revenue per ad platform

    Exactly these 3 — no "None of these" option (redundant on a multi-select; unchecked
    or Skip already covers it). Read references/blocks/<id>.md for each picked block.

  4. Build it — read references/live-dashboard.md and follow it exactly (tool-name
    resolution, template, placeholders, id/title). Config shape:

    { "version": 1, "domain": { "id": "uuid", "name": "..." }, "blocks": ["core-kpis", "purchase-funnel", "..."] }
    
  5. Confirm: "Store Pulse for [domain] is live in your sidebar — open it whenever you
    want to check that store."

  6. Offer scheduling (skip if already scheduled for this domain), then stop your turn:

    "Want me to set up a recurring report for this too? I can send it on whatever
    cadence works for you — email or Slack."

    Declined → end, don't re-offer this turn.


Edit an existing dashboard

Reached from "Save as artifact" step 2, or directly ("add channel performance to my
flyinmiata.com dashboard", "switch my atari.com dashboard to a different domain").

Read SP_CONFIG from the target artifact, mutate, rebuild via
references/live-dashboard.md, update_artifact on the same id — never create a
second artifact for a domain that already has one.

Every rebuild silently re-resolves the live noibu_search_sessions tool name from your
current session (never reuse the one baked into the old artifact — its instance id may
have changed) and passes it as mcp_tools. This self-heals stale tool names on every
edit — an "mcp_tools allowlist" error just means rebuild + update. Don't narrate this
resolution to the user.

  • Block change → update blocks, rebuild, update.
  • Domain change on this artifact:
    1. noibu_list_domains, match case-insensitively (accept www. variants); not found
      → list options, stop.
    2. Confirm: "Switching this dashboard from [current] to [new] — it'll replace this
      dashboard's data source. Continue?"
    3. Update SP_CONFIG.domain, rebuild, update. Id/title stay derived from the old
      domain unless asked to rename — cosmetic mismatch is harmless.
    4. Schedule exists for the old domain → update_scheduled_task (new domain baked in)
      and rename the task to store-pulse-[new-domain-slug].
    • Want a second dashboard for a different domain instead? → "Save as artifact" for
      that domain; leave this one untouched.
  • Artifact missing → its config went with it; re-run "Save as artifact".

Schedule report

Triggered by the snapshot widget's "Schedule report" button (arrives as
Schedule Store Pulse report for [domain]), a direct request, or step 6 above. Never
requires a saved artifact — the cron bakes its own config and never reads the live
dashboard.

  1. Check for an existing schedule. list_scheduled_tasks, look for
    store-pulse-[domain-slug] (see references/schedule.md's Task identity section) —
    found → this is an edit.

  2. Read references/schedule.md's ## Scheduling form widget and pass that HTML
    verbatim (the JS depends on every selector).

  3. Render via show_widget, title schedule_store_pulse — the only renderer that
    auto-sends on submit. Load read_me (modules: ["interactive"]) once, silently, if
    not already loaded.

  4. On submit (frequency, day, time, delivery, detail), resolve each picked
    delivery's connector before asking for its target — never fall back to a
    plain-text install link, use the connector-recommendation widget instead:

    • Email → look for a live Gmail (create_draft/search_threads/list_drafts)
      or Outlook draft tool.
      • Found → ask for the email address (ask which inbox first if both are available).
      • Neither found → search_mcp_registry(["gmail","email"]) and
        search_mcp_registry(["outlook","microsoft 365","email"]), then
        suggest_connectors(keywords:["email"], uuids: <both>) so Gmail and Microsoft
        365 show up together. Skip this delivery until connected; keep going with the
        rest.
    • Slack → look for a live slack_search_channels tool.
      • Found → ask which channel, resolve it.
      • Missing or unauthenticated → search_mcp_registry(["slack","messages","chat"]),
        then suggest_connectors(keywords:["messages"], uuids: <match>). Same
        skip-don't-block rule.
    • All deliveries unresolved → don't create the cron; tell them to come back once
      something's connected.
  5. Create/update the cronupdate_scheduled_task (existing) or
    create_scheduled_task name store-pulse-[domain-slug] (new). Bake
    frequency/day/time/detail plus only the resolved deliveries into the prompt
    template (references/schedule.md). Resolve and substitute the connector tool-name
    placeholders ({{slack_canvas_tool}}, {{slack_message_tool}},
    {{gmail_draft_tool}}, {{outlook_draft_tool}}) from your current session's tools —
    see references/schedule.md's "Connector tool names must also be baked in".

  6. Confirm, including the permission note:

    "Scheduled — your Store Pulse report for [domain] will go out [readable summary].
    One heads-up: you'll need to run the task once manually from your scheduled tasks
    to grant the permissions it needs. After that first run, it'll deliver on its own."

Skipped form → acknowledge, move on.


Migration note

Pre-multi-domain dashboards sit at the fixed id store-pulse-dashboard — leave exactly
as-is (no rename, recreate, or migrate). Any "does domain X already have a dashboard"
check must treat it as that domain's dashboard (match via SP_CONFIG.domain.name). Only
new dashboards use the domain-derived id scheme.


File map

store-pulse/
├── SKILL.md
├── references/blocks/{core-kpis,purchase-funnel,top-products,channel-performance,paid-performance}.md
├── references/inline-snapshot.md
├── references/live-dashboard.md
└── references/schedule.md
  • references/blocks/<id>.md — block specs (metrics, queries, tooltips); shared by the
    snapshot, the dashboard, and scheduled reports.
  • references/inline-snapshot.md — default response shape: summary, KPI+funnel widget,
    action buttons, investigate-thread question.
  • references/live-dashboard.md — persistent artifact: template, tool-name resolution,
    id/title convention.
  • references/schedule.md — cron windows, task naming, message formats, and the
    scheduling form widget.