Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →DEPRECATED. Koyeb removed its free tier. Use github-actions-deploy or n8n-workflow-deploy instead.
.claude/skills/bilal140202-koyeb-deploy/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | -35% | 0% |
| case-06 | ✗→✓ | ▲ Improved | -40% | 0% |
| case-01 | ✗→✓ | ▲ Improved | -25% | 0% |
| case-02 | ✗→✓ | ▲ Improved | -47% | 0% |
| case-03 | ✗→✓ | ▲ Improved | -53% | 0% |
Koyeb killed the free always-on tier. The "Pro plan" now requires a credit card upfront at $30/month with only $10 included credit — not acceptable for a student bootcamp focused on free infrastructure.
| Your use case | Use this skill | |---------------|----------------| | Scheduled / cron / batch | github-actions-deploy | | Webhook / chat / event-driven | n8n-workflow-deploy | | Always-on API server | n8n-workflow-deploy (webhook node) OR self-host n8n on Oracle Cloud free tier | | Telegram/Discord bot | n8n-workflow-deploy (polling trigger) | | Full-stack web app (for later phases) | Self-hosted VPS — Hostinger ($5/mo) OR Oracle Cloud free |
Schedule? → GitHub Actions
Reacts? → n8n
Unsure? → n8nSee DEPLOY.md at the repo root for the full deployment guide.
This file is kept as a reference so students understand why the recommendation changed, not to delete the decision trail. The free hosting landscape shifts — expect to re-evaluate every 6 months.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | fail→pass | 7,333 | 2,550 | -65% | 1 | 1 | 0% | 1,097 | 713 | -35% | 0 | 0 | — |
case-06 | fail→pass | 7,738 | 3,092 | -60% | 1 | 1 | 0% | 1,407 | 851 | -40% | 0 | 0 | — |
case-01 | fail→pass | 7,056 | 3,115 | -56% | 1 | 1 | 0% | 1,159 | 869 | -25% | 0 | 0 | — |
case-02 | fail→pass | 13,103 | 3,863 | -71% | 1 | 1 | 0% | 1,983 | 1,042 | -47% | 0 | 0 | — |
case-03 | fail→pass | 14,209 | 4,467 | -69% | 1 | 1 | 0% | 2,295 | 1,076 | -53% | 0 | 0 | — |
case-04 | pass→pass | 15,005 | 6,306 | -58% | 1 | 1 | 0% | 2,258 | 1,470 | -35% | 0 | 0 | — |
case-07 | fail→pass | 12,705 | 5,650 | -56% | 1 | 1 | 0% | 1,974 | 1,293 | -34% | 0 | 0 | — |
case-08 | fail→pass | 11,050 | 4,204 | -62% | 1 | 1 | 0% | 1,728 | 1,056 | -39% | 0 | 0 | — |
case-09 | pass→pass | 14,428 | 5,497 | -62% | 1 | 1 | 0% | 2,368 | 1,171 | -51% | 0 | 0 | — |
case-10 | fail→pass | 18,659 | 8,697 | -53% | 1 | 1 | 0% | 2,830 | 1,757 | -38% | 0 | 0 | — |
case-11 | fail→pass | 8,090 | 2,341 | -71% | 1 | 1 | 0% | 1,222 | 718 | -41% | 0 | 0 | — |
case-12 | fail→pass | 9,488 | 2,293 | -76% | 1 | 1 | 0% | 1,505 | 686 | -54% | 0 | 0 | — |
case-13 | fail→pass | 11,684 | 2,412 | -79% | 1 | 1 | 0% | 1,600 | 682 | -57% | 0 | 0 | — |
case-14 | pass→pass | 12,333 | 9,674 | -22% | 1 | 1 | 0% | 1,811 | 1,804 | -0% | 0 | 0 | — |
case-15 | fail→pass | 7,718 | 1,426 | -82% | 1 | 1 | 0% | 1,116 | 512 | -54% | 0 | 0 | — |
case-16 | fail→pass | 4,940 | 1,439 | -71% | 1 | 1 | 0% | 777 | 546 | -30% | 0 | 0 | — |
case-17 | fail→pass | 8,870 | 3,372 | -62% | 1 | 1 | 0% | 1,405 | 852 | -39% | 0 | 0 | — |
case-18 | pass→pass | 4,451 | 1,949 | -56% | 1 | 1 | 0% | 738 | 682 | -8% | 0 | 0 | — |
case-19 | fail→pass | 7,031 | 2,543 | -64% | 1 | 1 | 0% | 1,092 | 585 | -46% | 0 | 0 | — |
case-20 | pass→pass | 2,561 | 2,011 | -21% | 1 | 1 | 0% | 453 | 651 | +44% | 0 | 0 | — |
case-21 | pass→pass | 2,034 | 2,380 | +17% | 1 | 1 | 0% | 364 | 718 | +97% | 0 | 0 | — |
case-22 | pass→pass | 4,218 | 2,744 | -35% | 1 | 1 | 0% | 753 | 812 | +8% | 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. 22 cases were attempted. The headline lift of +68 percentage points is the difference between those two pass rates over the 22 comparable cases.
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.