Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add kyh/vibedgames --skill onboardinggit clone --depth 1 https://github.com/kyh/vibedgamesWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/kyh/vibedgames/onboarding)<a href="https://agentmods.dev/skills/kyh/vibedgames/onboarding"><img src="https://agentmods.dev/badge/skills/kyh/vibedgames/onboarding/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/kyh/vibedgames/onboarding"><img src="https://agentmods.dev/badge/skills/kyh/vibedgames/onboarding.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to medium
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- medium Excessive Agency · line 70 Skill enables autonomous high-impact decisions without human-in-the-loop verification. Critical operations (destructive commands, financial transactions, data deletion) should require explicit user confirmation.Fix: Add human-in-the-loop confirmation for destructive, irreversible, or high-impact operations. Never auto-execute commands that modify files, send data, or alter system state.
What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00033 | $0.01632 |
| Opus 5.5 | $0.00013 | $0.00653 |
| Sonnet 5.5 | $0.00007 | $0.00326 |
| Haiku 4.5 | $0.00003 | $0.00163 |
Grade A, and why
onboarding scanned grade A with 0 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 1mo ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 129 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Onboarding & difficulty
A vibedgames player arrives cold from a link, mid-conversation, with seconds of patience. The first session is the whole funnel. Fun is learning: the game lives while there's a pattern left to master, and difficulty must rise with skill to stay in the flow channel.
The first 30 seconds
- Time-to-first-input < 5s. No splash chain, no settings, no mode select.
- Time-to-core-verb < 30s. The player performs the signature verb (jump, shoot, plant) for real, with consequences, in the first half-minute.
- One input to start — "press any key / tap" over a live game scene. Show the actual game behind the title (attract mode), not a static logo.
- First input produces a juicy result: sound + motion + particles. The brain needs proof inputs matter before it will learn the pattern.
- Goal in one sentence or zero: "Survive." "Reach the flag."
- Detect input device: show touch hints to touch users, key hints to keyboard users — never a scheme the device can't use. If the core input is exotic (webcam, gamepad, mic), hint it explicitly and ship a plain mouse/keyboard fallback that's announced, not silent. Gimmick inputs (webcam/face/pose/hand, 3D presentation) are the product's identity — never drop them in a rebuild; standard inputs are fallbacks alongside. Hand tracking is undiscoverable on touch — hint it.
- Browser permission prompts are modals. Camera/mic/motion requests are
the most common cold-open modal in browser games — make them opt-in behind
a button, or defer until after first play. Never fire
getUserMediaon mount, and don't fetch the model before consent: the MediaPipe bundle pulled pre-consent from CDN is the biggest per-game load lever (6.1 MB → 0.15 MB and 2.5 MB → 0.14 MB on two games once deferred).
Teaching without a tutorial (George Fan's PvZ rules)
- Doing > reading. Never explain in text what a situation can force the player to discover once. Gate progress behind doing it once — done = taught.
- ≤ 8 words on screen at a time. If an instruction needs more, redesign the situation, not the sentence.
- Safe first exposure (Half-Life 2 pattern): observe the hazard eat a barrel before dodging it yourself.
- Ramp: teach → repeat with one twist → combine. Never two new ideas at once; roughly one new element per level.
- Objects self-describe: appearance telegraphs function; piggyback on real-world knowledge (red = damage, coins = money, arrows = direction).
- Hints are adaptive, not up-front: "Try planting further left!" only after the relevant failure. No modal popups that pause play.
- Bias the first decision: make the correct opening move the cheap, shiny, obvious one (Fan made sunflowers cheap and sparkly).
- Never label anything "Tutorial".
- Control hints live on the start screen only — never persistent chrome
during play (allowed exceptions in
game-ui).
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 1mo ago Changed · +7 lines · -90 tokens per session f8cd530a2e28
- 1mo ago First seen · 122 lines · 123 tokens per session scan A 8f64f17a047f
onboarding is a skill published in the GitHub repository kyh/vibedgames (61 stars, last pushed yesterday), licensed MIT. It adds 33 tokens to every session and 1,632 once invoked, about $0.0001 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
retro-learn
Convert delivery findings into skill, eval, workflow, and documentation improvements.
verify
Build/launch/drive recipe for verifying apps/docs changes at runtime (demos, docs pages, llms.txt).
quality-engineering-playwright-cli
Standardizes token-efficient browser automation via playwright-cli, with Playwright MCP as the fallback driver. Runs a preflight, picks the driver rung, captures aria snapshots/console/screenshots into a named evidence dir. Use for web verification, navigation, and capturing snapshots/logs.
system-design-case-catalog
Answer classic system design problems as constraint-to-solution sketches and coach interview practice: URL shortener, rate limiter, news feed, chat, notification, autocomplete, crawler, unique id. Use for interview practice or naming the closest known shape for a new problem.
system-design-interview-coaching
Facilitate timed system-design interview practice across backend, mobile/client, SDK/library, and full-stack surfaces. Use risk-driven probes, an adverse timeline, mechanism revision, one changed constraint, and the existing six-criterion evidence-based debrief.
quality-engineering-playwright-pom-generation
Generates Playwright page objects from an executable test plan, one class per screen with ladder-compliant locators and no assertions, wired into a shared fixture. Use when a web scenario needs a page object that does not exist yet, or when locators drift and the page object must be rebuilt.