arc-cost-reporting

name: arc-cost-reporting description: Daily cost and token usage report — tracks dual costs (Claude Code vs API estimates) updated: 2026-03-13 tags: - monitoring - cost - reporting

Sensor that generates a daily cost and token usage report. Runs every 60 minutes, creates one report per day. No thresholds or alerts — pure reporting.

Report Contents

Each report includes dual-cost breakdown (Claude Code + API estimate):

Daily totals — cost_usd (Claude Code), api_cost_usd (token-rate estimate), tokens, task count
Top tasks by cost — 5 most expensive tasks today (sorted by code cost, showing both cost fields)
Top tasks by tokens — 5 most token-intensive tasks today (tokens_in + tokens_out)
Top skills by cost — aggregated by the skills field; dual-cost breakdown per skill
Top sensors by cost — sensor-sourced tasks with dual-cost breakdown

How It Works

Runs every 60 minutes via claimSensorRun
One report per day (date-stamped source key prevents duplicates)
Queries local DB only
Report created as priority 9 / haiku task

When to Receive This Task

The report is informational. Review which skills or sensors are generating expensive work. If a particular skill is unexpectedly high, investigate whether its priority routing is correct (Opus vs Sonnet vs Haiku).

Cost Fields Explained

cost_usd: Actual Claude Code session consumption cost (what Anthropic charges)
api_cost_usd: Estimated API cost calculated from tokens × per-token rate (what the equivalent API calls would cost)

Both are tracked in tasks.cost_usd and tasks.api_cost_usd as documented in CLAUDE.md.

Checklist

skills/arc-cost-reporting/SKILL.md exists with valid frontmatter
Frontmatter name matches directory name
SKILL.md is under 2000 tokens
sensor.ts exports async default function returning Promise<string>
No threshold-based alerting — pure reporting
Dual cost metrics: cost_usd and api_cost_usd tracked in all report sections
Report breakdowns include both cost fields for comparison