Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Optimize Bright Data costs through tier selection, sampling, and usage monitoring. Use when analyzing Bright Data billing, reducing API costs, or implementing usage monitoring and budget alerts. Trigger with phrases like "brightdata cost", "brightdata billing", "reduce brightdata costs", "brightdata pricing", "brightdata expensive", "brightdata budget".
.claude/skills/jeremylongshore-brightdata-cost-tuning/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 109% | 0% |
| case-15 | ✗→✓ | ▲ Improved | 70% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 40% | 0% |
| case-18 | ✗→✓ | ▲ Improved | 27% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 47% | 0% |
Translate an approved workload into measurable provider and internal units, attribute them to an owner, and stop work before cost or scope escapes. Read current account and contract data at decision time; do not encode volatile prices in source.
Read usage evidence and Grep the implementation for product choice, attempts, retries, concurrency, bytes, browser time, snapshot records, downloads, delivery, storage, and downstream processing. Label provider-reported units separately from local estimates.
Write or Edit a usage envelope containing workload ID, owner, environment, product, target class, maximum attempts, maximum bytes or records, schedule, retention, and approved destination. Reject unattributed operations.
Calculate baseline, expected, and worst-authorized cases from current contract inputs. Model retries and duplicate processing explicitly, but do not assume a global provider request limit or a universal price.
Add preflight budgets, in-run alerts, hard abort thresholds, and post-run reconciliation against provider evidence. Investigate variance by product, target, failure class, byte volume, and duplicate work before raising a budget.
Use Read and Grep for usage, billing, and implementation inspection. Use Write and Edit for the usage envelope, forecast, alerts, tests, and reconciliation record. This skill does not change a plan, purchase capacity, or run collection traffic.
A batch has a maximum snapshot-record count, transfer-byte ceiling, retry budget, storage retention, and owner tag. The run stops at its authorized boundary and reconciles provider usage before the next schedule is approved.
| Failure | Meaning | Response | |---------|---------|----------| | Current contract inputs are unavailable | Forecast lacks an authority | Mark cost unknown and block expansion | | Usage has no workload owner | Spend is unattributed | Quarantine the schedule until ownership is assigned | | Variance comes from duplicate delivery | Processing is not idempotent | Fix deduplication before increasing the budget |
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-04 | pass→pass | 12,026 | 8,093 | -33% | 1 | 1 | 0% | 2,442 | 3,209 | +31% | 0 | 0 | — |
case-05 | fail→pass | 7,681 | 6,421 | -16% | 1 | 1 | 0% | 1,427 | 2,989 | +109% | 0 | 0 | — |
case-01 | fail→fail | 18,744 | 15,642 | -17% | 1 | 1 | 0% | 3,879 | 5,353 | +38% | 0 | 0 | — |
case-02 | fail→fail | 18,316 | 11,729 | -36% | 1 | 1 | 0% | 3,753 | 4,141 | +10% | 0 | 0 | — |
case-03 | fail→fail | 20,626 | 14,038 | -32% | 1 | 1 | 0% | 4,403 | 4,881 | +11% | 0 | 0 | — |
case-06 | pass→pass | 12,516 | 9,553 | -24% | 1 | 1 | 0% | 2,035 | 3,527 | +73% | 0 | 0 | — |
case-07 | pass→pass | 6,652 | 6,532 | -2% | 1 | 1 | 0% | 1,069 | 2,851 | +167% | 0 | 0 | — |
case-08 | pass→pass | 17,391 | 2,894 | -83% | 1 | 1 | 0% | 1,337 | 2,170 | +62% | 0 | 0 | — |
case-09 | pass→pass | 11,221 | 2,991 | -73% | 1 | 1 | 0% | 1,912 | 2,177 | +14% | 0 | 0 | — |
case-10 | pass→pass | 7,985 | 2,050 | -74% | 1 | 1 | 0% | 1,361 | 2,050 | +51% | 0 | 0 | — |
case-15 | fail→pass | 11,803 | 8,481 | -28% | 1 | 1 | 0% | 1,976 | 3,351 | +70% | 0 | 0 | — |
case-11 | pass→pass | 8,861 | 2,528 | -71% | 1 | 1 | 0% | 1,740 | 2,094 | +20% | 0 | 0 | — |
case-12 | pass→pass | 10,091 | 4,492 | -55% | 1 | 1 | 0% | 1,712 | 2,462 | +44% | 0 | 0 | — |
case-13 | fail→pass | 9,528 | 2,475 | -74% | 1 | 1 | 0% | 1,522 | 2,132 | +40% | 0 | 0 | — |
case-14 | pass→pass | 9,719 | 3,377 | -65% | 1 | 1 | 0% | 1,471 | 2,307 | +57% | 0 | 0 | — |
case-16 | pass→pass | 16,297 | 7,829 | -52% | 1 | 1 | 0% | 1,854 | 3,096 | +67% | 0 | 0 | — |
case-17 | pass→pass | 12,032 | 3,486 | -71% | 1 | 1 | 0% | 1,920 | 2,143 | +12% | 0 | 0 | — |
case-18 | fail→pass | 10,608 | 2,455 | -77% | 1 | 1 | 0% | 1,754 | 2,223 | +27% | 0 | 0 | — |
case-19 | fail→pass | 9,430 | 2,744 | -71% | 1 | 1 | 0% | 1,483 | 2,173 | +47% | 0 | 0 | — |
case-20 | pass→pass | 10,290 | 10,670 | +4% | 1 | 1 | 0% | 1,998 | 3,049 | +53% | 0 | 0 | — |
case-21 | pass→pass | 14,399 | 12,790 | -11% | 1 | 1 | 0% | 2,304 | 3,943 | +71% | 0 | 0 | — |
case-22 | pass→pass | 16,479 | 16,666 | +1% | 1 | 1 | 0% | 2,845 | 4,683 | +65% | 0 | 0 | — |
case-23 | fail→pass | 12,823 | 3,577 | -72% | 1 | 1 | 0% | 1,668 | 2,288 | +37% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 23 cases were attempted. The headline lift of +26 percentage points is the difference between those two pass rates over the 23 comparable cases.
The publisher has shipped newer versions since this run, so these numbers describe v1, not the version currently listed.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.