Multi Entity Forecast Dashboard

Multi-entity forecast dashboard

Produces one self-contained HTML file: Overview · Accounts · Forecast · Report tabs, an entity switcher (Group + each company file), scenario presets and sliders, and — on the MosoFin path — a live Sync from QuickBooks button and an Ask drawer that read the books through the connector.

The look is fixed. Read references/ui-spec.md before touching any HTML and never restyle ad hoc; the template already implements the spec.

Workflow

1. Choose the data path

Ask one question first: "Use your MosoFin connection to read the books, or supply the data yourself?" Then:

• MosoFin → follow references/mosofin-workflow.md (workspace confirm → company files → get_cash_flow per entity → get_aged_receivables → consolidate).
• Manual / opted out / no connector → follow references/manual-workflow.md (CSV schema, validation, what the dashboard hides).

Never mix: if a MosoFin read fails partway, tell the user what succeeded and offer the manual path for the rest rather than inventing figures.

2. Assemble the data object

Write a JSON file matching assets/example-data.json. Required keys: group, mode ("mosofin" | "manual"), source, fetched, year, months, order, entities, notes. Each entity needs label, short, and equal-length ocf, icf, fcf, ni arrays, plus ar, arIc, ap, inv; aging and customers are optional (null when not pulled).

notes drive every explanatory callout in the UI. Use where: "history" for run-rate exclusions, where: "accounts" for intercompany and aging caveats, where: "report" for sandbox/data-quality warnings. Write them from what the data actually shows — see references/forecast-method.md for what must be stated.

3. Build

python3 scripts/build_dashboard.py --data <data.json> --out /mnt/user-data/outputs/<group>-forecast-dashboard.html

or, for a CSV:

python3 scripts/build_dashboard.py --csv <file.csv> --group "<Group>" --mode manual --source "Uploaded CSV" --fetched "Uploaded <date>" --out <path>

4. Verify before presenting

Run the file headlessly (jsdom is fine) and confirm: zero JS errors, KPI band shows four values, the entity switcher lists Group + every entity, all four tabs render, and the first KPI equals the last cumulative cash figure you computed. Then present_files.

5. Tell the user

In prose, not bullets: the cash position, which scenario rate was chosen and why, the largest caveat (intercompany, sandbox data, missing aging), and — MosoFin path only — that the Sync and Ask buttons work when the file runs inside Claude with the connector attached and degrade to an amber strip elsewhere. End with a single "Data sources" line listing the tools and fetch time.

Files

Guardrails

• Every number in the dashboard comes from the tool results or the user's file. No interpolation, no "typical" values.
• Disclose sandbox signals and intercompany balances every time they appear.
• Read-only: this skill never writes to QuickBooks or MosoFin.
• Do not add chart libraries, fonts from CDNs, or new colors. If a requested change conflicts with the UI spec, say so and offer the closest spec-compliant option.