[ PACK LIBRARY / BUSINESS OPERATIONS ]

TASKFREE SAMPLE

Run a monthly Anthropic, Vercel, and Railway cost report

STAFF MILTON · V0.1.0
bopen-tools:cost-tracking

REQUIRES

  • claude plugin install bopen-tools@b-open-io

WHEN TO REACH FOR THIS

You want to know exactly what you spent on AI and infrastructure last month, broken down by model and service, without hand-reconciling three dashboards. Run this on a monthly cadence and infra/AI burn becomes a number you know before the invoice arrives, in a single sitting.

REQUIRES

  • claude plugin install bopen-tools@b-open-io (provides the cost-tracking skill)
  • An Anthropic Admin API key (sk-ant-admin..., org-admin scope) exported as ANTHROPIC_ADMIN_KEY
  • A Vercel API token with billing read access exported as VERCEL_TOKEN, plus your team slug or teamId
  • Railway has no billing API — you will read those figures from the Railway dashboard (Settings > Usage) and paste them in

CONTEXT INPUTS

{{REPORT_MONTH}}        # e.g. "2026-06" ; the calendar month to report
{{VERCEL_TEAM}}         # Vercel team slug or teamId
{{RAILWAY_USAGE}}       # paste the Railway dashboard usage total for the month, or "none"
{{BUDGET_CEILINGS}}     # optional: monthly budget per service, if you track them

THE PROMPT

PASTE INTO YOUR AGENT
You are Milton, the finance operator. Produce a monthly cost report across Anthropic,
Vercel, and Railway for {{REPORT_MONTH}}.

Load the bopen-tools:cost-tracking skill and follow it exactly. Then:

1. Anthropic: query the cost report endpoint for the full month at 1d granularity,
   and the usage report grouped by model. Report total USD spend from the cost
   report and a per-model breakdown from the usage report (tokens by model). The
   cost report groups by workspace and description rather than model, so any
   per-model dollar figure is a priced estimate; label it as such. Note explicitly
   that Priority Tier spend is NOT in the cost report if the org uses it.
2. Vercel: pull billing charges for {{VERCEL_TEAM}} for the month (FOCUS JSONL) and
   summarize spend by project.
3. Railway: do NOT attempt to fetch programmatically. Use the pasted {{RAILWAY_USAGE}}
   figure. If it is "none", state that Railway was not reported this month.
4. Compare each service total against {{BUDGET_CEILINGS}} if provided; flag any line
   over budget with the absolute and percentage delta.
5. Flag any single day where Anthropic spend exceeded 2x the 7-day rolling average,
   and name the model that drove it.

Deliver a single markdown report: a summary table (service, spend, budget, delta),
the per-model Anthropic breakdown, the Vercel per-project breakdown, anomalies, and
one paragraph of optimization recommendations (e.g. opus->sonnet downgrades for
low-complexity, high-volume calls).

Negative prompts: do not fabricate spend numbers; if an API call fails, report the
failure and which figure is missing. Do not attempt to fetch Railway costs from any
API. Do not claim a total is complete if Priority Tier is excluded; say so.

EXPECTED OUTPUT

A single markdown cost report for the month: summary table across all three services, per-model and per-project breakdowns, any budget overruns and daily spend anomalies flagged, and a short optimization note. Real numbers pulled live from the Anthropic and Vercel APIs; Railway filled from your pasted figure.

FOLLOW-UP CHAIN

  • bo-cost-dashboard — Build a real-time cost dashboard across Anthropic, Vercel, and Railway
  • bo-cost-investor-chain — Milton to Chief to Tina: cost anomaly to reprioritization to investor note
  • bo-financial-profile — Financial profile: budget ceilings, approval thresholds, cost centers

← PACK CONTENTS