Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when a val talks to an external service — Slack, Discord, Telegram, Stripe, GitHub, Gmail, Google Sheets, Postgres/Supabase/Upstash/Neon, browser automation (Playwright, Browserbase, Kernel, Steel), web scraping, PDF generation, push notifications, RSS, or any other third-party API. Covers the required workflow (fetch the Val Town guide, get credentials, test, store secrets) and the catalog of available guides.
.claude/skills/hashgraph-online-third-party-integrations/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 66% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 10% | 0% |
| case-07 | ✗→✓ | ▲ Improved | -2% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 22% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 25% | 0% |
When a val uses any external service, follow this order — do not skip steps and do not write integration code from training-data memory alone. Val Town's guides have platform-specific patterns and required workarounds that won't be in your training data.
https://docs.val.town/guides/, but the slug isn't always just the service name — some are grouped under a category (e.g. databases/neon-postgres/, browser-automation/kernel/) or live on a sub-page (slack/agent/). Don't guess the URL: fetch the docs sitemap at https://docs.val.town/sitemap-0.xml (it lists every docs URL), find the guides/… entry that matches the service, and fetch that page before writing any integration code.Deno.env.get("KEY_NAME") to read them, and document the required env vars in the README so the user (or anyone remixing the val) knows what to set. Whenever you reference an env var the user needs to set, show the raw, full URL to the prefilled Val Town env var editor on its own line, in this exact format: 👉 Add KEY_NAME here: https://www.val.town/x/HANDLE/VAL_NAME/environment-variables?key=KEY_NAME. Keep the URL visible (not hidden behind link text) — it's the call-to-action.Services with dedicated guides today — not exhaustive, so use the sitemap from step 1 as the current source of truth:
std/email instead — see the email skill)std/sqlite instead — see the sqlite-storage skill)std/oauth instead)Integration code is the most common place models hallucinate. APIs change, auth flows get reworked, and platform constraints (no filesystem, no subprocess) break naive approaches. The Val Town guide is the source of truth for what currently works on the platform.
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 12,552 | 23,072 | +84% | 1 | 1 | 0% | 2,442 | 4,302 | +76% | 0 | 0 | — |
case-02 | fail→pass | 18,163 | 25,623 | +41% | 1 | 1 | 0% | 3,604 | 5,972 | +66% | 0 | 0 | — |
case-03 | fail→fail | 20,142 | 29,506 | +46% | 1 | 1 | 0% | 2,758 | 3,705 | +34% | 0 | 0 | — |
case-04 | pass→pass | 8,539 | 13,309 | +56% | 1 | 1 | 0% | 1,517 | 2,133 | +41% | 0 | 0 | — |
case-05 | pass→pass | 15,523 | 19,293 | +24% | 1 | 1 | 0% | 1,769 | 3,280 | +85% | 0 | 0 | — |
case-06 | fail→pass | 22,862 | 21,849 | -4% | 1 | 1 | 0% | 3,470 | 3,829 | +10% | 0 | 0 | — |
case-07 | fail→pass | 11,446 | 12,647 | +10% | 1 | 1 | 0% | 1,931 | 1,886 | -2% | 0 | 0 | — |
case-08 | fail→pass | 13,424 | 11,872 | -12% | 1 | 1 | 0% | 2,261 | 2,768 | +22% | 0 | 0 | — |
case-09 | fail→pass | 10,194 | 14,574 | +43% | 1 | 1 | 0% | 1,963 | 2,456 | +25% | 0 | 0 | — |
case-10 | fail→fail | 7,719 | 15,085 | +95% | 1 | 1 | 0% | 1,577 | 1,567 | -1% | 0 | 0 | — |
case-11 | pass→pass | 13,924 | 23,009 | +65% | 1 | 1 | 0% | 2,280 | 3,508 | +54% | 0 | 0 | — |
case-12 | fail→pass | 15,165 | 11,971 | -21% | 1 | 1 | 0% | 3,040 | 3,034 | -0% | 0 | 0 | — |
case-13 | fail→fail | 16,646 | 23,374 | +40% | 1 | 1 | 0% | 2,068 | 2,547 | +23% | 0 | 0 | — |
case-14 | fail→pass | 24,436 | 26,460 | +8% | 1 | 1 | 0% | 3,423 | 3,373 | -1% | 0 | 0 | — |
case-15 | fail→pass | 7,373 | 9,527 | +29% | 1 | 1 | 0% | 1,279 | 1,606 | +26% | 0 | 0 | — |
case-16 | fail→pass | 11,357 | 21,348 | +88% | 1 | 1 | 0% | 2,203 | 3,736 | +70% | 0 | 0 | — |
case-17 | fail→fail | 25,573 | 28,454 | +11% | 1 | 1 | 0% | 3,988 | 4,262 | +7% | 0 | 0 | — |
case-18 | fail→pass | 16,163 | 7,335 | -55% | 1 | 1 | 0% | 2,676 | 1,810 | -32% | 0 | 0 | — |
case-19 | fail→pass | 22,623 | 32,165 | +42% | 1 | 1 | 0% | 2,666 | 2,361 | -11% | 0 | 0 | — |
case-20 | fail→pass | 10,191 | 8,140 | -20% | 1 | 1 | 0% | 1,682 | 1,227 | -27% | 0 | 0 | — |
case-21 | pass→pass | 23,363 | 24,347 | +4% | 1 | 1 | 0% | 3,461 | 3,633 | +5% | 0 | 0 | — |
case-22 | fail→pass | 13,736 | 5,348 | -61% | 1 | 1 | 0% | 2,563 | 1,821 | -29% | 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, and 21 counted toward the lift figure. The other 1 produced results that are not comparable between the two arms, so they are excluded from the headline rather than averaged into it. The headline lift of +59 percentage points is the difference between those two pass rates over the 21 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.