{"id":51100791,"url":"https://github.com/raydocs/fusion-deck","last_synced_at":"2026-06-24T10:30:54.356Z","repository":{"id":364750504,"uuid":"1269083006","full_name":"raydocs/fusion-deck","owner":"raydocs","description":"🃏 Three models gang up and out-argue the lone star. A Claude Code skill: a panel of Opus 4.8 + GPT-5.5 + Gemini 3.1 Pro judged into one answer — plus plan, context, orchestrate \u0026 handoff workflows. Runs on the subscriptions already on your machine.","archived":false,"fork":false,"pushed_at":"2026-06-14T10:07:40.000Z","size":3795,"stargazers_count":1,"open_issues_count":0,"forks_count":0,"subscribers_count":0,"default_branch":"main","last_synced_at":"2026-06-14T11:22:12.124Z","etag":null,"topics":["ai-agents","anthropic","claude","claude-code","code-review","developer-tools","ensemble","fusion","gemini","llm","multi-model","prompt-engineering"],"latest_commit_sha":null,"homepage":null,"language":"Shell","has_issues":true,"has_wiki":null,"has_pages":null,"mirror_url":null,"source_name":null,"license":"mit","status":null,"scm":"git","pull_requests_enabled":true,"icon_url":"https://github.com/raydocs.png","metadata":{"files":{"readme":"README.md","changelog":null,"contributing":null,"funding":null,"license":"LICENSE","code_of_conduct":null,"threat_model":null,"audit":null,"citation":null,"codeowners":null,"security":null,"support":null,"governance":null,"roadmap":null,"authors":null,"dei":null,"publiccode":null,"codemeta":null,"zenodo":null,"notice":null,"maintainers":null,"copyright":null,"agents":null,"dco":null,"cla":null}},"created_at":"2026-06-14T09:37:22.000Z","updated_at":"2026-06-14T10:13:29.000Z","dependencies_parsed_at":null,"dependency_job_id":null,"html_url":"https://github.com/raydocs/fusion-deck","commit_stats":null,"previous_names":["raydocs/fusion-deck"],"tags_count":null,"template":false,"template_full_name":null,"purl":"pkg:github/raydocs/fusion-deck","repository_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/raydocs%2Ffusion-deck","tags_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/raydocs%2Ffusion-deck/tags","releases_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/raydocs%2Ffusion-deck/releases","manifests_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/raydocs%2Ffusion-deck/manifests","owner_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners/raydocs","download_url":"https://codeload.github.com/raydocs/fusion-deck/tar.gz/refs/heads/main","sbom_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories/raydocs%2Ffusion-deck/sbom","scorecard":null,"host":{"name":"GitHub","url":"https://github.com","kind":"github","repositories_count":286080680,"owners_count":34726613,"icon_url":"https://github.com/github.png","version":null,"created_at":"2022-05-30T11:31:42.601Z","updated_at":"2026-05-26T15:22:16.424Z","status":"online","status_checked_at":"2026-06-24T02:00:07.484Z","response_time":106,"last_error":null,"robots_txt_status":"success","robots_txt_updated_at":"2025-07-24T06:49:26.215Z","robots_txt_url":"https://github.com/robots.txt","online":true,"can_crawl_api":true,"host_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub","repositories_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repositories","repository_names_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/repository_names","owners_url":"https://repos.ecosyste.ms/api/v1/hosts/GitHub/owners"}},"keywords":["ai-agents","anthropic","claude","claude-code","code-review","developer-tools","ensemble","fusion","gemini","llm","multi-model","prompt-engineering"],"created_at":"2026-06-24T10:30:53.729Z","updated_at":"2026-06-24T10:30:54.349Z","avatar_url":"https://github.com/raydocs.png","language":"Shell","funding_links":[],"categories":[],"sub_categories":[],"readme":"# fusion-deck\n\n\u003cp align=\"center\"\u003e\n  \u003cimg src=\"assets/hero.png\" alt=\"fusion-deck — a panel of models, one judged answer\" width=\"100%\"\u003e\n\u003c/p\u003e\n\n\u003e 🃏 Three B-tier models gang up and out-argue the one A-tier star.\n\u003e A Claude Code skill that turns a panel of models into one judged answer — plus a workflow toolkit that\n\u003e plans, investigates, gathers context, splits, optimizes, refactors, and hands off.\n\n![Claude Code Skill](https://img.shields.io/badge/Claude%20Code-Skill-8A63D2)\n![License: MIT](https://img.shields.io/badge/License-MIT-3da639.svg)\n\n**English** · [简体中文](#简体中文)\n\n---\n\n## The story\n\nOpenRouter published a fun result: a **panel of models, judged by one of them, beats the best single\nfrontier model** (“Fusion beats frontier”). Two snags with just using theirs:\n\n1. The single strongest model in that test — **Claude Fable 5 — is off the table for me. I can't run it.**\n2. OpenRouter's Fusion is a **metered API**: every call costs.\n\nSo fusion-deck does the same trick, **on your own machine**: it rounds up **three models you already pay a\nflat subscription for** — Claude Opus 4.8, GPT‑5.5 (via the `codex` CLI), and Gemini 3.1 Pro (via\nAntigravity CLI `agy`; legacy `gemini` is opt-in) — has Opus 4.8 judge them, and **beats the lone star\nanyway**. No extra per‑token API meter:\nit just rides the CLIs you're already logged into. Three cobblers, one Zhuge Liang. 🧠\n\n```mermaid\nflowchart LR\n    Q([\"Your question\"]) --\u003e O[\"Opus 4.8\"]\n    Q --\u003e G[\"GPT-5.5\"]\n    Q --\u003e M[\"Gemini 3.1 Pro\"]\n    O --\u003e J{{\"Opus 4.8 judges\u003cbr/\u003econsensus · conflicts · blind spots\"}}\n    G --\u003e J\n    M --\u003e J\n    J --\u003e R([\"One cross-checked answer\"])\n```\n\n\u003e **The catch:** the full panel needs all three subscriptions/CLIs. Missing one? No drama — it runs with\n\u003e whatever you've got and always tells you exactly which models answered.\n\n## The proof\n\nOpenRouter's **DRACO** deep‑research benchmark — 100 tasks across 10 domains:\n\n| Setup | DRACO | vs. best solo model |\n| --- | --- | --- |\n| 🃏 **fusion-deck's panel** — Opus 4.8 + GPT‑5.5 + Gemini 3.1 Pro, judged by Opus 4.8 | **68.3%** | **+3.0** 🟢 |\n| Opus 4.8 + GPT‑5.5, judged by Opus 4.8 | 67.6% | +2.3 |\n| 🌟 Claude Fable 5 — the lone star, solo | 65.3% | — _(baseline)_ |\n| GPT‑5.5, solo | 60.0% | −5.3 |\n| Opus 4.8, solo | 58.8% | −6.5 |\n\nThe three underdogs land **68.3% — that's +3.0 over the star (Fable 5) and ~+9.5 over Opus 4.8 on its\nown.** Three independent tries catch each other's mistakes; even the *same* model run twice and judged\njumps +6.7. Not luck — that's the whole point.\n\n*Data: OpenRouter, “[Fusion beats frontier](https://openrouter.ai/blog/announcements/fusion-beats-frontier/).”\nfusion-deck runs the same panel locally via Claude, `codex`, and `agy` — no router, nothing\nleaves for a third party.*\n\n## Two superpowers\n\n**① Think hard — open the panel.**\n`/fusion \u003cquestion\u003e` and `/fusion-review \u003ccode or diff\u003e` fan your question (or your code) out to the panel,\nblind and in parallel, then Opus 4.8 judges it into **one cross‑checked answer** — or one prioritized\nfindings list, must‑fix first. For the calls where being confidently wrong is expensive.\n\n**② Work smart — run the workflow.** This is the part people sleep on:\n\n- 🧩 **`/fusion-plan \u003cone fuzzy line\u003e`** → a real plan: the goal, a concrete “done‑when”, the steps, the\n  risks. Stop hand‑holding the AI through vague asks — pin down what you actually meant first.\n- 📦 **`/fusion-context \u003ctask\u003e`** → a tidy, **token‑budgeted context pack of only the files that matter**.\n  The model finally reasons about your real code instead of drowning in the whole repo.\n- 🔀 **`/fusion-orchestrate \u003ctask\u003e`** → **splits the work into pieces, runs each in a focused sub‑agent, and\n  verifies each one before starting the next.** Big changes done carefully — not one hopeful mega‑prompt.\n- 🔎 **`/fusion-investigate \u003cbug or \"why is it like this\"\u003e`** → evidence first, then the panel adjudicates\n  the competing theories. A root‑cause report, not a confident guess.\n- ⏱️ **`/fusion-optimize \u003cmetric\u003e`** → a measure → change → re‑measure loop: baseline first, one change at a\n  time, the panel calls continue/stop. No baseline, no bragging.\n- ♻️ **`/fusion-refactor \u003ctarget\u003e`** → structure analysis → behavior‑preserving plan → one steered agent.\n  Cleaner code, same behavior (proven by the tests that stay green).\n- 🤝 **`/fusion-handoff \u003cwork\u003e`** → a clean handoff note (done / verified / risks / next steps) so the next\n  agent — or future‑you — picks up in seconds.\n\n**Power-user modes:** `/fusion-plan --deep` (a polished design doc with a critique pass) · `/fusion-context\n--discover` (let an agent curate the pack, evidence-gated) · `/fusion-orchestrate --worktrees` (isolate\nparallel siblings in their own git worktrees). All opt-in; the plain commands stay simple.\n\nChain them and you go from a vague one‑liner to a verified, shipped change:\n\n```text\nfuzzy idea → /fusion-plan → /fusion-context → /fusion-orchestrate → /fusion-handoff\n```\n\nUnder the hood it's tuned to actually *get* you: panelists answer **blind** (no echo chamber), the judge\n**reconciles** consensus vs. contradictions (it doesn't average), context is **curated not dumped**, and\nevery step is **verified before the next**.\n\n## Which command? (when to use what)\n\nNot sure which to reach for? Match your situation below — and when the task is easy, just ask Claude\ndirectly; the panel is for the calls where being wrong is expensive.\n\n| When you're trying to… | Reach for | Panel? |\n| --- | --- | --- |\n| Settle a hard call or trade-off (*\"optimistic or pessimistic locking?\"*) | `/fusion` | yes |\n| Vet code, a diff, or a plan before it ships | `/fusion-review` | yes |\n| Find the root cause of a bug, or *\"why is it built like this?\"* | `/fusion-investigate` | by exception |\n| Turn a vague idea into a concrete, checkable plan | `/fusion-plan` · `--deep` for a design doc | no |\n| Hand the *right* files to another model or agent | `/fusion-context` · `--discover` to auto-curate | no |\n| Execute a big, multi-step change carefully | `/fusion-orchestrate` · `--worktrees` to parallelize | no |\n| Make something measurably faster or smaller | `/fusion-optimize` | by exception |\n| Clean up structure **without** changing behavior | `/fusion-refactor` | no |\n| Pass work to the next agent (or future-you) | `/fusion-handoff` | no |\n| Re-anchor a drifting session (situation→command + invariants) | `/fusion-remind` | no |\n\nTypical flows: a **feature** is `plan → context → orchestrate → handoff`; a **bug** is\n`investigate → plan → orchestrate`. A one-off hard question is just `/fusion`.\n\n## Install\n\n```bash\ngit clone https://github.com/raydocs/fusion-deck.git\nbash fusion-deck/install.sh\n```\n\nThen run **`/reload-skills`** in Claude Code (or restart). Done — `/fusion`, `/fusion-plan`, … are ready.\n\n**For the full 3‑model panel**, install the two optional CLIs (and be logged into each):\n\n- [`codex`](https://developers.openai.com/codex) — adds the GPT‑5.5 panelist\n- [`agy`](https://antigravity.google/docs/cli-install) — adds the Gemini 3.1 Pro panelist via Antigravity CLI\n  - Legacy `gemini` is still available only when explicitly enabled with `FUSION_GEMINI_BACKEND=gemini`\n    or `FUSION_ALLOW_LEGACY_GEMINI=1`.\n\nCheck anytime:\n\n```bash\nbash ~/.claude/skills/fusion-deck/scripts/detect_panel.sh   # which models are available right now\nbash ~/.claude/skills/fusion-deck/scripts/smoke_test.sh     # offline self-check (never calls a paid model)\n```\n\n## Examples\n\n```text\n/fusion Should we use optimistic or pessimistic locking for the booking flow? Trade-offs at our scale.\n/fusion-review git diff main...HEAD\n/fusion-investigate the cart total is wrong for multi-currency orders\n/fusion-plan add a /health endpoint with a test\n/fusion-context the checkout flow, so I can hand it to another agent\n/fusion-orchestrate docs/plans/add-health.md\n/fusion-optimize cut p95 latency of /search under load; stop at 200ms\n/fusion-refactor the payments module\n/fusion-handoff the auth refactor\n```\n\n## Good to know\n\n- **Where the savings come from.** It reuses the subscriptions you're already logged into (Claude /\n  `codex` / Antigravity `agy`) — no per‑token API bill the way OpenRouter's Fusion API charges. *You just\n  need the three subscriptions.* The full panel costs more quota and runs as slow as its slowest model, so only\n  `/fusion` and `/fusion-review` open the whole table by default, `/fusion-investigate` and\n  `/fusion-optimize` call it only at their decision points, and the rest are fast single‑model commands.\n- **Nothing is faked.** Every panel answer states which models actually answered; a smaller panel is never\n  dressed up as the full one.\n- **No secrets in the repo.** Auth lives in the CLIs; nothing private is hardcoded.\n\n## License\n\n[MIT](LICENSE)\n\n---\n\n## 简体中文\n\n\u003e 🃏 三个臭皮匠合起来，比那个独苗状元还能打——这回状元叫 Fable。\n\u003e 一个 Claude Code 技能：把一桌模型拧成一个被评审过的答案，外加一套会规划、会查根因、会备上下文、会拆活、会调优、会重构、会交接的工作流。\n\n[English](#fusion-deck) · **简体中文**\n\n### 来历\n\nOpenRouter 发了个挺好玩的结论：**一桌模型 + 其中一个当评审，分数能压过最强的单个前沿模型**（《Fusion beats frontier》）。但直接用他们的有俩坎：\n\n1. 那场里最能打的单模型 —— **Claude Fable 5，我这儿根本用不了，被封了。**\n2. OpenRouter 的 Fusion 是 **按量计费的 API**：一调一掏钱。\n\n所以 fusion-deck 把这套搬到**你自己电脑上**：拉上**三个你本来就按月订阅、早就登录好的模型** —— Claude Opus 4.8、GPT‑5.5（走 `codex`）、Gemini 3.1 Pro（默认走 Antigravity CLI `agy`，旧 `gemini` 只做显式兼容）—— 让 Opus 4.8 当评审，**照样把那个单飞的状元比下去**。不额外按 token 收费，直接复用你已经登录的订阅。三个臭皮匠，顶个诸葛亮。🧠\n\n```mermaid\nflowchart LR\n    Q([\"你的问题\"]) --\u003e O[\"Opus 4.8\"]\n    Q --\u003e G[\"GPT-5.5\"]\n    Q --\u003e M[\"Gemini 3.1 Pro\"]\n    O --\u003e J{{\"Opus 4.8 评审\u003cbr/\u003e共识 · 冲突 · 盲点\"}}\n    G --\u003e J\n    M --\u003e J\n    J --\u003e R([\"一个交叉核对过的答案\"])\n```\n\n\u003e **小前提：** 想凑齐整桌，你得有这三家的订阅 / CLI。少一个也不耽误 —— 它会用现有的接着跑，而且每次都老老实实告诉你这回到底上了谁。\n\n### 实测\n\nOpenRouter 的 **DRACO** 深度研究基准 —— 10 个领域、100 道题：\n\n| 配置 | DRACO | 比最强单模型 |\n| --- | --- | --- |\n| 🃏 **fusion-deck 的阵容** —— Opus 4.8 + GPT‑5.5 + Gemini 3.1 Pro，Opus 4.8 评审 | **68.3%** | **+3.0** 🟢 |\n| Opus 4.8 + GPT‑5.5，Opus 4.8 评审 | 67.6% | +2.3 |\n| 🌟 Claude Fable 5 —— 独苗状元，单飞 | 65.3% | —（基准） |\n| GPT‑5.5，单飞 | 60.0% | −5.3 |\n| Opus 4.8，单飞 | 58.8% | −6.5 |\n\n三个臭皮匠落在 **68.3% —— 比状元 Fable 5（65.3%）高 3.0 分**，比 Opus 4.8 单飞高了将近 9.5 分。三次各自独立的尝试会互相挑错；哪怕同一个模型跑两遍再合并，也能高 6.7 分。不是运气，这就是整件事的核心。\n\n*数据来自 OpenRouter 的《[Fusion beats frontier](https://openrouter.ai/blog/announcements/fusion-beats-frontier/)》（DRACO 基准）。fusion-deck 是用你本机的 Claude / `codex` / `agy` 直接跑同一套阵容 —— 不经过任何 router，也不往第三方发东西。*\n\n### 两样看家本领\n\n**① 想得狠 —— 开一桌。**\n`/fusion \u003c问题\u003e`、`/fusion-review \u003c代码 / diff\u003e`：把问题（或代码）甩给一桌模型，各自盲答、并行跑，再由 Opus 4.8 评审成**一个交叉核对过的答案** —— 或者一份排好优先级、必改的排最前的问题清单。专治\"答错了很贵\"的场合。\n\n**② 干得巧 —— 跑工作流。** 这部分最容易被低估：\n\n- 🧩 **`/fusion-plan \u003c一句模糊的话\u003e`** → 一份真计划：目标、怎样算做完、分几步、有哪些坑。别再手把手哄着 AI 猜你想要啥 —— 先把你真正的意思钉死。\n- 📦 **`/fusion-context \u003c任务\u003e`** → 一份卡着 token 预算、**只装该看的文件**的上下文包。让模型对着你真正的代码动脑子，而不是被整个仓库淹死。\n- 🔀 **`/fusion-orchestrate \u003c任务\u003e`** → **把活拆成小块，每块交给一个专注的子 agent，做完一块先验过再开下一块。** 大改动也能稳稳落地，而不是赌一个超长 prompt 一把梭。\n- 🔎 **`/fusion-investigate \u003cbug 或\"这玩意儿为啥长这样\"\u003e`** → 先把证据摆清楚，再让一桌模型给互相打架的几个假设当裁判。最后给你一份能指到根因的报告，而不是一拍脑袋的猜测。\n- ⏱️ **`/fusion-optimize \u003c指标\u003e`** → 量一下 → 改一处 → 再量一遍的循环：先立基线，一次只动一处，该接着干还是收手让一桌模型拍板。没基线，就不准吹优化。\n- ♻️ **`/fusion-refactor \u003c目标\u003e`** → 先看结构哪儿乱、哪儿重复，再排一份\"只动结构、不动行为\"的计划，然后盯着一个 agent 一步步落地。代码更干净，行为照旧——测试从头到尾绿着，就是没改坏的凭据。\n- 🤝 **`/fusion-handoff \u003c工作\u003e`** → 一份干净的交接（做了啥 / 验了啥 / 有啥风险 / 下一步），下一个 agent —— 或者明天的你 —— 接手秒上手。\n\n**进阶玩法：** `/fusion-plan --deep`（产出一份正式设计文档，中途还会自己挑一遍刺）· `/fusion-context --discover`（让 agent 自己挑文件，但每个都得拿得出证据）· `/fusion-orchestrate --worktrees`（并行的几路各跑在自己的 git worktree 里，互不踩脚）。都是可选项，平时用基础命令照样省心。\n\n串起来用，一句模糊需求就能走到一个验证过、能交付的改动：\n\n```text\n模糊想法 → /fusion-plan → /fusion-context → /fusion-orchestrate → /fusion-handoff\n```\n\n底层都是冲着\"更懂你\"调的：几个模型**盲答**（不搞回声室）、评审**分清共识和冲突**（不是求平均）、上下文**精挑而非乱塞**、每一步**验过再走**。\n\n### 用哪个？（什么时候用什么）\n\n拿不准用哪个？对着下面找你的处境就行——简单活儿直接问 Claude，一桌模型是留给\"答错了很贵\"的场合的。\n\n| 你想干的 | 用 | 开整桌？ |\n| --- | --- | --- |\n| 拍一个难决定 / 权衡（*\"乐观锁还是悲观锁？\"*） | `/fusion` | 是 |\n| 上线前审一段代码 / diff / 方案 | `/fusion-review` | 是 |\n| 查一个 bug 的根因，或*\"这玩意儿为啥长这样？\"* | `/fusion-investigate` | 按需 |\n| 把一句模糊想法变成能落地、能验收的计划 | `/fusion-plan` · `--deep` 出设计文档 | 否 |\n| 把**该看的**文件挑给另一个模型 / agent | `/fusion-context` · `--discover` 自动挑 | 否 |\n| 稳稳执行一个多步的大改动 | `/fusion-orchestrate` · `--worktrees` 并行 | 否 |\n| 想把啥改得更快 / 更小（数字看得见） | `/fusion-optimize` | 按需 |\n| 只整理结构、**不改**行为 | `/fusion-refactor` | 否 |\n| 把活交给下一个 agent（或明天的你） | `/fusion-handoff` | 否 |\n| 长会话跑偏了，或新 agent 要一眼看清地图和铁律 | `/fusion-remind` | 否 |\n\n常见流程：**功能** = `plan → context → orchestrate → handoff`；**改 bug** = `investigate → plan → orchestrate`。临时一个难问题，直接 `/fusion`。\n\n### 安装\n\n```bash\ngit clone https://github.com/raydocs/fusion-deck.git\nbash fusion-deck/install.sh\n```\n\n然后在 Claude Code 里跑一下 **`/reload-skills`**（或者直接重启），就齐活了 —— `/fusion`、`/fusion-plan`…… 拿来就能用。\n\n**想凑齐三个模型的完整阵容**，再装两个可选 CLI（并各自登录好）：\n\n- [`codex`](https://developers.openai.com/codex) —— 接上 GPT‑5.5\n- [`agy`](https://antigravity.google/docs/cli-install) —— 通过 Antigravity CLI 接上 Gemini 3.1 Pro\n  - 旧 `gemini` 只在显式设置 `FUSION_GEMINI_BACKEND=gemini` 或 `FUSION_ALLOW_LEGACY_GEMINI=1` 时启用。\n\n想随时检查一下：\n\n```bash\nbash ~/.claude/skills/fusion-deck/scripts/detect_panel.sh   # 现在能用上哪几个模型\nbash ~/.claude/skills/fusion-deck/scripts/smoke_test.sh     # 本地自检（不花钱、不碰付费模型）\n```\n\n### 来几个例子\n\n```text\n/fusion 预订流程到底用乐观锁还是悲观锁？按我们这个量级帮我权衡下\n/fusion-review git diff main...HEAD\n/fusion-investigate 多币种订单的购物车总价算错了\n/fusion-plan 加一个带测试的 /health 接口\n/fusion-context 把结账流程整理一下，我要交给另一个 agent\n/fusion-orchestrate docs/plans/add-health.md\n/fusion-optimize 把 /search 的 p95 延迟压下来，目标 200ms\n/fusion-refactor 支付模块\n/fusion-handoff 这次的鉴权重构\n```\n\n### 几点说明\n\n- **省钱省在哪。** 它复用你电脑里已经登录的订阅（Claude / `codex` / Antigravity `agy`），不像 OpenRouter Fusion 那样按 token 收 API 费 —— **前提是你有这三家的订阅。** 整桌一起上更费额度、也得等最慢的那个，所以默认只有 `/fusion` 和 `/fusion-review` 开整桌，`/fusion-investigate` 和 `/fusion-optimize` 只在关键决策点才开桌，其余命令走单模型，图个快。\n- **不糊弄。** 每个面板答案都会写明这回到底是哪几个模型回答的；小阵容绝不冒充满配。\n- **仓库里不放密钥。** 登录的事交给各家 CLI，绝不往代码里塞私密信息。\n\n### 许可证\n\n[MIT](LICENSE)\n","project_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fraydocs%2Ffusion-deck","html_url":"https://awesome.ecosyste.ms/projects/github.com%2Fraydocs%2Ffusion-deck","lists_url":"https://awesome.ecosyste.ms/api/v1/projects/github.com%2Fraydocs%2Ffusion-deck/lists"}