deepep-to-cam-converter

deepep-to-cam-converter is a skill for Claude Code, Codex from XiaoLuoLYG/GOD. It costs 63 tokens per session (3,040 once invoked), scanned A, original, Apache-2.0.

A code migration skill for replacing DeepEP communication operations with CAM operations when moving mixture-of-experts code from CUDA/NCCL to Ascend NPUs.

In plain words
What is it for?
Use it to inspect DeepEP and distributed-training code, adapt CUDA and NCCL-specific parts, and replace supported operations.
Why use it?
It helps identify incompatible operations and check whether the real runtime parameters meet the target hardware's constraints.

Skill for Claude CodeCodex

Written for no agent in particular: nothing here depends on one.

Good fit Use it to inspect DeepEP and distributed-training code, adapt CUDA and NCCL-specific parts, and replace supported operations.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/xiaoluolyg/god/deepep-to-cam-converter
About the project

GOD is a control room for observing and directing societies of language-model agents running in simulated worlds. It lets researchers inspect replays, question individual agents, alter future events, reset simulations, and export experiments for reuse. The catalogue entries are skills and agents for operating and investigating these simulations.

XiaoLuoLYG/GOD · 1,107 stars · on GitHub · xiaoluolyg.github.io

Install

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.

Any agent
npx skills add XiaoLuoLYG/GOD --skill deepep-to-cam-converter
Clone the repo
git clone --depth 1 https://github.com/XiaoLuoLYG/GOD

Made for: Claude Code, Codex.

Wrote 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.

agentmods badge for deepep-to-cam-converter

README.md
[![agentmods](https://agentmods.dev/badge/skills/xiaoluolyg/god/deepep-to-cam-converter/github.svg)](https://agentmods.dev/skills/xiaoluolyg/god/deepep-to-cam-converter)
Your own site
<a href="https://agentmods.dev/skills/xiaoluolyg/god/deepep-to-cam-converter"><img src="https://agentmods.dev/badge/skills/xiaoluolyg/god/deepep-to-cam-converter/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.

agentmods 80×15 button for deepep-to-cam-converter

Your own site · 80×15
<a href="https://agentmods.dev/skills/xiaoluolyg/god/deepep-to-cam-converter"><img src="https://agentmods.dev/badge/skills/xiaoluolyg/god/deepep-to-cam-converter.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 63 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 3,040 The whole file, excluding the scripts and references it only reads on demand.
Security scan A 0 findings. A grade says what 26 rules found in the file — not that it is safe. Third-party audits
  • NVIDIA SkillSpector pass 7 Sept 2026
How audits are shown
Origin original No closer match found in the catalogue.
Token cost

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.

ModelPer sessionOnce invoked
Fable 5.1 $0.00063 $0.03040
Opus 5 $0.00032 $0.01520
Sonnet 5 $0.00013 $0.00608
Haiku 4.5 $0.00006 $0.00304

Measured 8d ago against content hash 4e729708335c, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-11, from the pricing page.

Security

Grade A, and why

deepep-to-cam-converter 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 8d 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.

jiuwenclaw/jiuwenclaw/resources/agent/jiuwenclaw_workspace/skills/deepep-to-cam-converter/SKILL.md · 168 lines

How it starts

The opening of the file, as written. The whole thing — 168 lines — stays where its author put it; the contents beside it link to each section on GitHub.

CAM算子替换专家技能

技能概述

本技能专注于将基于DeepEP的Mixture of Experts (MoE) 代码迁移至昇腾(Ascend)NPU环境。它能自动识别DeepEP dispatch & combine算子,**严格基于代码的实际运行参数(而非默认值)**校验CAM算子约束。

核心原则

  • 思维链先行:在执行任何代码修改前,必须先输出分析过程和自检结果。严禁直接生成代码。
  • 原地修改:默认直接在用户指定的原文件上修改,严禁创建新文件(除非用户显式要求)。
  • 动态值优先:判断约束时,必须追踪参数的实际运行时值,严禁直接使用代码中的硬编码默认值。
  • 强制交互:在算子模式选择(如A3普通 vs Shmem)和功能支持度处理上,必须询问用户,严禁AI自作主张。

工作流程

1. 代码扫描与依赖分析

1.1 目标检测
  • DeepEP识别:扫描代码中是否存在import deep_epdispatch/combine相关调用。
  • MOE/EP模式识别:确认代码是否涉及Mixture of Experts或Expert Parallelism逻辑。
  • 无匹配处理:若未检测到相关代码,直接告知用户并终止流程。
1.2 依赖与上下文分析
  • 导入分析:提取所有import语句,识别本地模块依赖。
  • 通信域扫描:检查是否存在dist.init_process_groupbackend='nccl'torch.cuda.set_device等特征代码。
  • 外部依赖检查:识别是否调用了外部初始化函数(如utils.init_dist),需同步分析这些函数的实现。

🛑 阶段 1 阻断检查(必须显式输出) 在生成任何回复前,请先输出以下内容:

[阶段 1 自检]
- DeepEP 代码识别:[是/否]
- 本地依赖文件列表:[列出文件]
- 状态:[通过/失败]

若“DeepEP 代码识别”为否,直接终止。若为是,继续下一阶段。

2. 环境与策略决策

注意:此阶段必须与用户交互,不得跳过。

2.1 环境选择

向用户展示以下选项,确定目标运行环境:

  • 选项1:基于当前环境自动检测(尝试执行npu-smi info,根据设备型号判断是A2环境或A3环境)。
  • 选项2:指定目标环境(用户手动指定A2环境或A3环境,用于交叉编译或代码预研场景)。
2.2 参数值解析与约束校验

重要原则:代码中的默认值(Default Value)≠ 实际运行值(Runtime Value)

在执行替换前,必须对约束条件涉及的参数进行实际值溯源

  1. 参数溯源

    • 检查参数是否通过argparseconfig.yaml或命令行传入。
    • 严禁使用 parser.add_argument(..., default=XXX) 中的 XXX 作为判断依据。
    • 必须假设代码中的默认值可能不是实际运行值。
  2. 交互确认机制

    • 对于关键约束参数(如 hidden_size, top_k, expert_num 等),如果无法从代码逻辑中100%确定其运行时赋值,必须向用户发起询问:

      “检测到参数 [参数名] 在代码中的默认值为 [默认值],但这可能不是实际运行值。请确认实际运行时的数值是多少?以便准确判断是否满足CAM算子约束。”

  3. 约束校验逻辑

    • 情况A(值确定且满足):用户确认实际值,且满足CAM算子约束 -> 进入算子模式选择
    • 情况B(值确定但不满足):用户确认实际值,不满足约束 -> 停止替换,告知用户不支持。
    • 情况C(值不确定):用户未提供 -> 暂停替换,标记该处需人工确认。
2.3 算子模式选择

在确认参数值满足约束后,根据目标环境确定具体算子模式:

  • A2环境:仅使用 CAM A2 算子
  • A3环境
    • 检查代码是否同时满足 普通A3算子Shmem算子fused deep moe算子的约束条件。
    • 注意fused deep moe算子替换有严格的约束范式,请仔细阅读说明文档,检查是否满足范式,需要基于dispatch & combine函数去查找调用链,查看最终调用方是否满足约束范式
    • 阻断规则如果有多个都满足,AI必须立即停止自动流程,并向用户展示选择题,等待用户回复,以三个条件都满足的场景为例,可以展示如下选项

      检测到当前代码同时满足多组算子的约束条件,请做出选择:()

      • 选项1(普通A3):参数配置简单,无需额外共享内存初始化。
      • 选项2(Shmem A3):性能更优,但需满足特定约束及初始化。
      • 选项2(fused deep moe算子):性能最优,将通信计算过程融合为一个大算子。 请回复选项继续。
    • 如果仅满足其一:自动选择该模式并告知用户。

Read the full file on GitHub · 168 lines

Files

What ships with it

5 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.

Changes

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.

  1. 8d ago First seen · 168 lines · 63 tokens per session scan A 4e729708335c

Subscribe to this mod's changes

deepep-to-cam-converter is a skill published in the GitHub repository XiaoLuoLYG/GOD (1,107 stars, last pushed 15d ago), licensed Apache-2.0. It adds 63 tokens to every session and 3,040 once invoked, about $0.0003 per session on Opus 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 2026-09-03.

Related

Other skills, from other repositories

camsnap

Capture frames or clips from RTSP/ONVIF cameras. Grabs snapshots, video clips, and motion events from IP cameras, security cameras, and video streams. Use when the user wants to take a snapshot from a camera, record a clip from an RTSP stream, monitor motion on a security camera, discover ONVIF devices on the network…

elizaOS/eliza · 84 tokens

blind-sqli

Blind SQL injection under hostile WAF — manual bypass playbook for when sqlmap fails because common tokens (SUBSTRING, IF, AND, WHERE, single quotes) are filtered. Covers token-fingerprinting probe loops, arithmetic-multiplication boolean evaluation, hex-encoded literals, and exponential-probe binary search. Loaded on…

PurpleAILAB/Decepticon · 88 tokens

race-condition

Race condition / TOCTOU exploitation — concurrent and parallel-request attacks against web applications that check then act, write session state before validating it, or perform slow operations that widen the race window. Covers single-endpoint races (double-spend, coupon abuse, balance overflow) and multi-endpoint…

PurpleAILAB/Decepticon · 80 tokens

wikipedia

Search and read Wikipedia via x wkp — MediaWiki API, no API key, zero install; query, extract, suggest, and DDG route in one module. Load for wiki, wikipedia, encyclopedia lookup, article summary.

x-cmd/x-cmd · 49 tokens

api-server-sent-events

Server-Sent Events (SSE / EventSource) exploitation — origin abuse for cross-site streaming exfil, prompt-injection via SSE messages into LLM clients, retry-after token leak, fragmenting events to bypass content-type sniffers.

PurpleAILAB/Decepticon · 54 tokens

golang-grpc

Provides gRPC usage guidelines, protobuf organization, and production-ready patterns for Golang microservices. Use when implementing, reviewing, or debugging gRPC servers/clients, writing proto files, setting up interceptors, handling gRPC errors with status codes, configuring TLS/mTLS, testing with bufconn, or…

samber/cc-skills-golang · 72 tokens