Imported from jayhjenkins/ProductOSv0.2 (.claude/skills/metric-scorecard-fetch/SKILL.md). Install upstream with npx skills add jayhjenkins/ProductOSv0.2 --skill metric-scorecard-fetch. Copyright stays with the author.

Scorecard Fetch & Render (Dispatcher)

This skill is the long-lived scorecard orchestrator. It holds only metric slugs + results — never all metric specs at once. Each metric's calculation lives in its own self-contained file under .claude/skills/metric-scorecard-fetch/metrics/<slug>.md. These definitions are independent of the temporal quarterly Rocks/OKR skill — do not couple them.

Flow

1. Resolve as_of_date = last completed Saturday on/before today: python3 -c "import datetime as d;t=d.date.today();print(t-d.timedelta(days=(t.weekday()-5)%7))"

2. Read datasets/scorecard/registry.json to get the metric list.

3. Fetch home-wau first (the /1k-WAU metrics and stickiness need its value). Dispatch a subagent (general-purpose) handed ONLY the contents of metrics/home-wau.md + the as_of_date.

4. Dispatch the remaining auto metrics in parallel — one subagent each, in a single message with multiple Agent calls. Hand each subagent ONLY:

   • the contents of its own metrics/<slug>.md
   • the resolved as_of_date
   • if its file lists depends_on: home-wau, also the already-computed home-wau integer value Instruct each to execute its spec exactly and return ONLY its JSON contract line (nothing else).

5. Handle non-auto metrics:

   • manual (hoai-weekly-interactions): reuse the prior week's value from values.json if present (status manual_carried); else value null, status manual.
   • deferred (new-resident-activation-rate): value null, status deferred.

6. Collect all results into a dict keyed by slug, write it with the Write tool to /tmp/scorecard_results_<as_of>.json, then record + render: python3 scripts/build_scorecard_dashboard.py --week <as_of> --record /tmp/scorecard_results_<as_of>.json

7. Report the dashboard path and a one-line summary per metric (value + status). Flag any error/stale metric and any value outside its file's stated Sanity range.

Conventions

• Pendo subId=4818486697721856. Databricks is_prod, READ-ONLY (never modify data).
• Saturday-anchored as_of; Sun–Sat weeks; windows inclusive of as_of.
• Never overwrite past weeks — the renderer appends to the time series.
• If a subagent errors, record status=error (carry last-known value if any) and continue; never let one metric block the others.
• Do NOT write to SharePoint — the human copies values from the rendered HTML.