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 xi-zhao/OpenQuantum --skill merlin-learninggit clone --depth 1 https://github.com/xi-zhao/OpenQuantumWrote 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/xi-zhao/openquantum/merlin-learning)<a href="https://agentmods.dev/skills/xi-zhao/openquantum/merlin-learning"><img src="https://agentmods.dev/badge/skills/xi-zhao/openquantum/merlin-learning/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/xi-zhao/openquantum/merlin-learning"><img src="https://agentmods.dev/badge/skills/xi-zhao/openquantum/merlin-learning.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 | $0.00041 | $0.00758 |
| Opus 5.5 | $0.00016 | $0.00303 |
| Sonnet 5.5 | $0.00008 | $0.00152 |
| Haiku 4.5 | $0.00004 | $0.00076 |
Grade A, and why
merlin-learning 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.
What it actually says
MerLin 光子层推断与梯度
原生 SDK 路径
本能力同时支持 MerLin 的原生公开 API。用户需要 自定义光学电路、QuantumLayer、测量策略与 Torch 训练 时，由 Agent 在固定依赖环境中编写和运行任务代码，不限于下文便捷 Tool 的预设模型。
使用 node scripts/sdk-access.mjs inspect merlin-learning 定位环境，按完整 SDK 工作流执行。模块导航：merlin、perceval；导航不是 API 白名单。
下文关于模型、门集、输入与预算的限制描述便捷 Tool 的合同；原生 SDK 另受上游版本、可选依赖、宿主资源与用户授权约束。已禁用或被拒绝的动作不能借代码路径绕过。执行状态和科研验收分别报告。
Agent 经 Harness 调用 merlin_local 的 evaluate_merlin_layer。用于分析光子量子层的可微响应；完整训练模型、准确率对比和实机推断需要其他工作流与证据。
输入模式数、Fock occupation、目标 targetOccupation、BS/PS 操作序列及二维 phases。每个 phase 行是一次参数取值，各列按线路中 PS 出现的顺序绑定。BS.theta 使用 Perceval Rx 约定：对角元 cos(theta/2)，非对角元 i sin(theta/2)；所有角度为弧度。目标与输入光子总数必须相同；MerLin SLOS 后端要求输入至少一个光子。每条线路至少有一个 PS，参数可跨批次比较。
Tool 显式使用完整 FOCK 空间而非 MerLin 默认的 UNBUNCHED 空间，保留光子聚束事件且不进行后选择。返回 basis 明确列出概率列对应的模式占据数；phaseGradients 是对应批次的目标输出概率对各 PS 角度的偏导，采用 CPU float64 自动微分。检查 probabilitySums，并用有限差分检验关键参数点；目标梯度不代表模型精度提升或量子优势。模拟假定光子不可区分、理想无损器件，无有限 shots 统计噪声。
固定 merlinquantum 0.4.1（MIT）与 Perceval 1.2.1，避免较新 Perceval 的不兼容模块变更。先运行 node scripts/setup-paper-tools.mjs merlin-learning；调用不安装依赖、不连接云、不加载任意用户模型。SDK 缓存使最大副作用登记为 workspace-write。execution 控制部署资源，模式、光子、批次数不设测试规模上限。
当前为 L1，scientificValidation=not_evaluated。开发期 permanent 参照和梯度回归不构成科研结论或最终科学验收。
What ships with it
6 files beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
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 · +11 lines 541848435fe1
- 18d ago First seen · 17 lines · 41 tokens per session scan A 99ea8d2dcb64
merlin-learning is a skill published in the GitHub repository xi-zhao/OpenQuantum (75 stars, last pushed yesterday), licensed MIT. It adds 41 tokens to every session and 758 once invoked, about $0.0002 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 2026-09-29.
Other skills, from other repositories
academic-research
Search academic papers, scholarly articles, and research publications through SandBase. Use when asked for literature review, academic citations, scholarly research, paper discovery, or scientific evidence gathering.
rhylthyme
Schedule multi-step procedures that a person carries out against a clock, with parallel tracks, shared equipment and steps that end when the person says so. Use when turning a lab protocol, recipe, event run sheet or workout into a timed plan; when several procedures must share one instrument (one centrifuge, one…
memory-search-mcp
Long-term memory and context retrieval for AI agents. Semantic search across conversation history with care-weighted relevance.
gc-minimal-zine-poster-v0-3
Generate or analyze poetic paper-texture minimal zine posters with large negative space, a small editorial collage or focal element, experimental typography, and one clear color accent. Use when the user gives a theme, sentence, article, mood, object, photo, content brief, reference images, or an image folder and…
ask-sonner
Guide to Sonner, the React toast library — install and wire up the Toaster, pick the right toast() call, promise and loading toasts, updating, dismissing and persisting toasts, styling, theming and icons, positioning and multiple toasters. Use when working with Sonner or troubleshooting it — toasts that don't appear…
kill-ai-slop
Find and remove AI slop — the generic, machine-default visual and copy tics of vibe-coded products — from a web project. Use when the user asks to "kill AI slop", "de-slop", "remove the AI look", "make this not look AI-generated", or clean up a landing page / UI / docs that feels templated. Detects and fixes the…