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 SunflowersLwtech/sunkit --skill frontier-modelsgit clone --depth 1 https://github.com/SunflowersLwtech/sunkitWrote 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/sunflowerslwtech/sunkit/frontier-models)<a href="https://agentmods.dev/skills/sunflowerslwtech/sunkit/frontier-models"><img src="https://agentmods.dev/badge/skills/sunflowerslwtech/sunkit/frontier-models/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/sunflowerslwtech/sunkit/frontier-models"><img src="https://agentmods.dev/badge/skills/sunflowerslwtech/sunkit/frontier-models.svg" alt="Reviewed on agentmods" width="80" height="20"></a>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 | $0.00181 | $0.01490 |
| Opus 5.5 | $0.00072 | $0.00596 |
| Sonnet 5.5 | $0.00036 | $0.00298 |
| Haiku 4.5 | $0.00018 | $0.00149 |
Grade A, and why
frontier-models 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 4d 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 — 48 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Frontier models and harnesses
Plain claude talks to Anthropic on this machine's Claude seat. Everything else goes through one of two doors.
1. The magpie gateway: other models inside Claude Code
claude-gw starts a Claude Code session on the gateway (llm-gateway, http://100.116.218.86:3425; this machine's key is in ~/.claude/magpie-settings.json). The /model picker in a claude-gw session lists the frontier rows below; /model <id> takes any id harness-run models prints.
Routing groups (magpie routing=order, stays=session: one member per session for the prompt cache, the next member when it fails or runs out):
| id | chain |
|---|---|
claude-opus-5-5 (Opus row, claude-gw's default) |
Opus 5.5 on the three Claude Team seats → Kiro Opus 5.5 → claude-sonnet-5-5 |
claude-sonnet-5-5 |
Sonnet 5.5 on the seats → Zed Sonnet 5.5 → Kiro Sonnet 5.5 → frontier |
claude-haiku-5-5 |
Haiku 5.5 on the seats → Zed Haiku 5.5 → Kiro Haiku 5.5 → DeepSeek V4.1 Flash → GLM-5.3 Flash → Go Haiku |
claude-fable-5-1 |
claude-opus-5-5 (no Fable on the gateway) |
frontier |
GPT-6.1 Sol (Zed) → Grok 4.7 → GPT-6 Sol → GPT-6.1 Sol (Trae) → Kimi K3.1 Pro → GLM-5.3 → DeepSeek V4 Pro → Qwen3.8 Max → MiniMax M3 → Doubao Seed 2.1 Pro → the OpenCode Go copies |
auto/<model> (magpie finds these) |
one model on whichever subscription renews soonest and has quota, e.g. auto/gpt-6.1-sol, auto/glm-5.3, auto/kimi-k3.1-pro, auto/deepseek-v4-pro, auto/qwen3.8-max |
Subscriptions behind it: Kiro Student (1000 credits a month), Zed Student, Grok (SuperGrok), WorkBuddy free, Trae CN and Trae Global, OpenCode Go, Copilot Student (only copilot/auto; named models are refused on the Student plan). The Claude seats are served by magpie's subscription bridge (a Claude Code process on the server per conversation), each capped at 95% of its five hours. A session stays on one seat; when that seat is used up it moves to the next seat, and the move retells the whole conversation to a new process once, which costs a full cache write (up to 835K tokens for a long session; on the first night one seat's five hours went in under 40 minutes this way). So the gateway keeps conversations of about 200K tokens or more off the seats: they go to the same chain without the seats (long-opus, long-sonnet, long-haiku), and wait there when it is all out of allowance. Never point an already-running session at the gateway, and prefer a new claude-gw session to resuming a long one.
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.
- 4d ago Changed 927896178acc
- 5d ago First seen · 48 lines · 181 tokens per session scan A 77ee9a52af41
frontier-models is a skill published in the GitHub repository SunflowersLwtech/sunkit (13 stars, last pushed yesterday), licensed Apache-2.0. It adds 181 tokens to every session and 1,490 once invoked, about $0.0007 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 2026-10-25.
Other skills, from other repositories
tidyfactor-styler
Production framework styler and surgical RTL UI polish engine with Contextual Decision Layer (CDL). Operates directly inside existing codebases across React/Next.js, PHP, WordPress, and Vanilla HTML/CSS/JS. Trigger on commands 'brief', 'component', 'section', 'page', 'redesign', 'typography', 'palette', 'layout'…
ui-styles
Participate before building or changing product UI: websites, apps, demos, dashboards, forms, visual redesigns, component-library selection, themes and interaction states. Act as a practical designer: understand the product, recommend and preview a direction, preserve a project visual standard, and implement it using…
ui-style-creator
A tool for creating and saving a reusable visual style from a product brief, an existing interface, or reference designs. A visual style can include shared choices used across an interface or project.
lovable-design-system
Distinctive, production-grade frontend design with bold aesthetic direction, curated palette/typography/layout presets, and a structured design-direction flow. Use whenever building or refining UI (landing pages, product surfaces, marketing sites, dashboards) — especially when the brief is visual, broad, or "make it…
material-ui-tailwind
Integrates Material UI with Tailwind CSS v4 using cascade layers (enableCssLayer, @layer order) and documents Tailwind v3 interoperability (preflight, important, injectFirst, portals). Use when combining MUI with Tailwind utilities, slotProps className, or theme token bridges.
fast-typescript-check
Keep www-sacred's TypeScript fast to type-check and fast to run. Use when touching the ASCII/canvas animation components (the only real per-frame code here), tightening type-check wall-clock, or auditing a change for runtime or compiler regressions. Scoped to this repo — a React 19 / Next.js 16 component library plus…