메뉴
Burn-rate trend over time with optional drift-alert exit code. Bins session spend into buckets, surfaces window-over-window delta, and can exit 1 when latest bucket exceeds prior mean by a configurable %. Distinct from `cost-trend` (benchmark drift); this tracks PRODUCTION spend trajectory.
| name | cost-burn |
| description | Burn-rate trend over time with optional drift-alert exit code. Bins session spend into buckets, surfaces window-over-window delta, and can exit 1 when latest bucket exceeds prior mean by a configurable %. Distinct from `cost-trend` (benchmark drift); this tracks PRODUCTION spend trajectory. |
| argument-hint | [--bucket 1d] [--lookback 14d] [--alert-on-acceleration-pct 50] [--format table|json] |
| allowed-tools | Bash |
Burn-rate trend observability. The fourth leg of the cost-tracker forward-cost stack:
| Question | Skill |
|---|---|
| "Have we crossed a threshold?" (reactive) | cost-budget-check |
| "When will we cross a threshold?" (predictive) | cost-projection |
| "Could we have spent less?" (comparative) | cost-counterfactual |
| "Is daily burn ACCELERATING?" (trend) | cost-burn ← this |
Implementation: scripts/burn.mjs.
session-* records from cost-tracking namespace.--bucket duration windows (default 1d) over --lookback (default 14d).{n: sessions, spendUsd: sum(total_cost_usd)}.delta = latest.spendUsd - mean(prior non-empty buckets).--alert-on-acceleration-pct N is set: exit 1 when deltaPct > N.| Latest bucket spend | $0.500000 (1 sessions) |
| Prior bucket mean | $0.100000 (4 non-empty buckets) |
| **Delta (latest vs prior mean)** | **+$0.400000 (400.00%)** |
# | Window | Sessions | Spend
0 | 2026-06-15 14:16 → 2026-06-16 14:16 | 1 | $0.500000
1 | 2026-06-14 14:16 → 2026-06-15 14:16 | 0 | $0.000000
2 | 2026-06-13 14:16 → 2026-06-14 14:16 | 1 | $0.100000
3 | 2026-06-12 14:16 → 2026-06-13 14:16 | 1 | $0.100000
...
$ cost burn --bucket 1d --lookback 7d --alert-on-acceleration-pct 50 ⚠ ALERT: latest bucket $0.500000 is 400.0% above prior mean $0.100000 (threshold +50%) exit 1
$ cost burn --bucket 1d --lookback 7d --alert-on-acceleration-pct 500 ✓ latest bucket within +500% of prior mean (actual delta: 400.0%) — OK exit 0
# Fail the build if today's spend accelerated > 100% over the weekly mean
cost burn --bucket 1d --lookback 7d --alert-on-acceleration-pct 100 || alert-oncall
The alert is independent of budget — it triggers on rate ACCELERATION even when total spend is well under budget. Catches "we shipped a hot loop that burns 10× normal" before the budget alarm goes off.
cost-trend| Skill | Data source | Question |
|---|---|---|
cost-trend | docs/benchmarks/runs/*.json | "Is the benchmark drifting (win rate, latency)?" |
cost-burn | cost-tracking namespace | "Is production spend accelerating?" |
Both useful; they answer different questions on different data.
Infinity/null in JSON, marked new in the table. No alert fires (no baseline to compare against).--bucket > --lookback: hard error, exit 2.MAD-based outlier detection on session spend. Robust to the very outliers it hunts (unlike mean+sigma). Surfaces specific anomalous sessions with modified-z scores; optional --alert-on-outliers exit code for CI gates. Distinct from cost-burn (aggregate trend) — this answers "which INDIVIDUAL session is the outlier?".
Multi-baseline counterfactual cost analysis. Compares actual session spend to hypothetical always-haiku / always-sonnet / always-opus routing baselines. Answers "is the routing earning its keep?" Negative savings flag over-escalation; positive savings quantify the router's win.
Snapshot delta between two cost-summary JSON outputs. PR-level cost regression detection — answers "what changed between these two specific snapshots?". Pairs with cost-summary's stable JSON contract.
Composite CI gate — runs cost-budget-check + cost-burn + cost-anomaly + cost-projection in parallel and surfaces a single combined health status with max exit code. The operationally-useful entry point — one shell-out covers all four alert ladders.
Forward-looking spend extrapolation. Computes a USD-per-day rate from the recent measurement window, projects to 7d/30d/90d/365d horizons, and surfaces "days until budget exhausted" when a budget is configured. Predictive counterpart to `cost-budget-check` (reactive).
Per-message cost breakdown within a single session. The drill-down companion to cost-anomaly — when an outlier session is flagged, this surfaces the specific expensive messages so operators can see whether the cost came from output tokens, cache writes, or model escalations.