yangwhale
UserMulti-platform AI bot framework: 5 hot-swappable agent runtimes (Claude Code / OpenClaw / Kilo / Gemini CLI / DeepSeek Harness) across Discord, Feishu, DingTalk and a self-hosted web entrance, with shared memory, bot-to-bot collaboration and real-time voice.
Categories
Indexed Skills (50)
launch-nemo-rl
Playbook for launching, monitoring, stopping, and debugging NeMo-RL recipes on a Kubernetes cluster via the nrl-k8s CLI. Covers ephemeral vs long-lived RayCluster modes, iterating on runs, and debugging hung or failed training jobs.
mcore-create-issue
Investigate a failing GitHub Actions run or job and create a GitHub issue for the failure.
mcore-run-on-slurm
How to launch distributed Megatron-LM training jobs on a SLURM cluster. Covers a minimal sbatch skeleton, environment-variable setup for torch.distributed.run, CUDA_DEVICE_MAX_CONNECTIONS rules across hardware and parallelism modes, container conventions, monitoring, and per-rank failure diagnosis.
mcore-split-pr
Split a PR into multiple PRs to reduce the number of required CODEOWNERS reviewer groups.
mcore-testing
Test system for Megatron-LM. Covers test layout, recipe YAML structure, adding and running unit and functional tests, golden values, marker filters, and CI parity.
nemo-automodel-distributed-training
Guide for selecting and configuring distributed training strategies in NeMo AutoModel, including FSDP2, Megatron FSDP, DDP, and parallelism settings.
nemo-automodel-launcher-config
Configure NeMo AutoModel job launches for interactive runs, Slurm clusters, and SkyPilot cloud execution.
nemo-automodel-model-onboarding
Guide for onboarding new model architectures into NeMo AutoModel, including architecture discovery, implementation patterns, registration, and validation.
nemo-automodel-recipe-development
Create and modify NeMo AutoModel training and evaluation recipes, including YAML structure, builders, and execution flow.
tilegym-converting-cutile-to-triton
Converts cuTile GPU kernels (@ct.kernel) to Triton (@triton.jit). Handles standard in-repo conversion, debugging (cudaErrorIllegalAddress, shape mismatch, numerical mismatch), and mapping cuTile idioms (ct.load/ct.store, ct.Constant, ct.launch) to Triton equivalents. Covers dual-kernel layout flags (e.g. transpose=True/False + autotune grid via META) per translations/advanced-patterns.md. Use when converting, porting, or translating cuTile kernels to Triton, or debugging existing Triton translations.
tilegym-cutile-python
Expert cuTile programming assistant. Write high-performance GPU kernels using cuTile's tile-based programming model with proper validation and optimization. Supports deep agent orchestration for complex multi-kernel tasks.
tilegym-improve-cutile-kernel-perf
Iteratively optimize cuTile kernel performance through systematic profiling, bottleneck analysis, IR comparison, and targeted tuning. Covers tile sizes, occupancy, autotune configs, TMA, latency hints, persistent scheduling, num_ctas, flush_to_zero, and IR-level debugging. Use when asked to "optimize cutile kernel", "improve kernel perf", "tune cutile performance", "make kernel faster", or iteratively benchmark and refine a cuTile GPU kernel in the TileGym project.
tilegym-monkey-patch-kernels-to-transformers
Integrate TileGym kernels into Hugging Face `transformers` models by replacing the library's submodule(s) and certain class(es)' implementations, and patching certain class(es)' init/forward/load weight methods prior to instantiating models. Used when the user requires integrating TileGym kernels into `transformers` models.
mcore-linting-and-formatting
Linting and formatting for Megatron-LM. Covers running autoformat.sh, tools (ruff, black, isort, pylint, mypy), and code style rules.
nemo-mbridge-mlm-bridge-training
Run Megatron-LM (MLM) and Megatron Bridge training with mock or real data. Covers correlation testing, available recipes, and multi-GPU examples.
nemo-mbridge-multi-node-slurm
Convert single-node scripts to multi-node Slurm sbatch jobs and debug common multi-node failures. Covers srun-native vs uv run torch.distributed approaches, container setup, NCCL timeouts, OOM sizing for MoE models, and interactive allocation.
nemo-mbridge-perf-activation-recompute
Validate and use selective and full activation recompute in Megatron Bridge to reduce GPU memory usage at the cost of extra compute.
nemo-mbridge-perf-cpu-offloading
Validate and use CPU offloading in Megatron Bridge, including layer-level activation offloading and fractional optimizer state offloading with HybridDeviceOptimizer.
nemo-mbridge-perf-cuda-graphs
Validate and use CUDA graph capture in Megatron Bridge, including local full-iteration graphs and Transformer Engine scoped graphs for attention, MLP, and MoE modules.
nemo-mbridge-perf-expert-parallel-overlap
Validate and use MoE expert-parallel communication overlap in Megatron-Bridge, including overlap_moe_expert_parallel_comm, delay_wgrad_compute, and flex dispatcher backends such as DeepEP and HybridEP.
nemo-mbridge-perf-hierarchical-context-parallel
Operational guide for enabling hierarchical context parallelism in Megatron-Bridge, including config knobs, code anchors, pitfalls, and verification.
nemo-mbridge-perf-megatron-fsdp
Operational guide for enabling Megatron FSDP in Megatron-Bridge, including config knobs, code anchors, pitfalls, and verification.
nemo-mbridge-perf-memory-tuning
Techniques for reducing peak GPU memory in Megatron Bridge — expandable segments, parallelism resizing, activation recompute, CPU offloading constraints, and common OOM fixes.
nemo-mbridge-perf-moe-comm-overlap
MoE expert-parallel communication overlap in Megatron Bridge. Covers dispatch/combine overlap, flex dispatcher backends, and expert wgrad scheduling.
nemo-mbridge-perf-moe-dispatcher-selection
Choose the right MoE token dispatcher (`alltoall`, DeepEP, or HybridEP) for the hardware, EP degree, and optimization stage. Summarizes patterns from DSV3, Qwen3, Qwen3-Next, and VLM bring-up work.
nemo-mbridge-perf-moe-hardware-configs
Representative MoE training playbooks by hardware platform and model family. Summarizes rounded throughput bands, parallelism patterns, and common tuning stacks.
nemo-mbridge-perf-moe-long-context
Long-context MoE training guidance for Megatron Bridge. Covers CP sizing, selective recompute, dispatcher choices, and practical patterns from DSV3, Qwen3, and Qwen3-Next long-context experiments.
nemo-mbridge-perf-moe-optimization-workflow
Systematic workflow for MoE training optimization in Megatron Bridge, based on the Megatron-Core MoE paper. Covers the Three Walls framework, parallel folding, recompute strategy, dispatcher choice, and CUDA-graph bring-up.
nemo-mbridge-perf-moe-vlm-training
Practical guidance for training MoE VLMs in Megatron Bridge. Compares FSDP and 3D-parallel approaches, using rounded lessons from Qwen3-VL, Qwen3-Next, and other multimodal experiments.
nemo-mbridge-perf-parallelism-strategies
Operational guide for choosing and combining parallelism strategies in Megatron Bridge, including sizing rules, hardware topology mapping, and combined parallelism configuration.
nemo-mbridge-perf-sequence-packing
Validate and use packed sequences and long-context training in Megatron-Bridge, distinguishing offline packed SFT for LLMs from in-batch packing for VLMs, and applying the right CP constraints.
nemo-mbridge-perf-tp-dp-comm-overlap
Operational guide for enabling TP, DP, and PP communication overlap in Megatron-Bridge, including config knobs, code anchors, pitfalls, and verification.
nemo-mbridge-recipe-recommender
Recommend and customize Megatron Bridge recipes for a user's model, GPU count, and training goal. Indexes library recipes (pretrain/SFT/PEFT) and performance recipes.
nemo-mbridge-resiliency
Resiliency features in Megatron Bridge including fault tolerance, straggler detection, in-process restart, preemption, and re-run state machine.
nemo-rl-auto-research
Autonomous NeMo-RL research agent workflow for directed hypothesis testing and open-ended discovery. Guides agents through the full experiment lifecycle: understanding recipes and environments, wiring RL or NeMo-gym runs, launching reproducible baselines and iterations, analyzing results, preserving human oversight, and using git plus TSV logs as the research ledger. Do NOT use for: bug fixes, code review, documentation, refactoring, dependency updates, or single-file changes.
nemo-rl-brev-etiquette
Brev instance operating guidance for NeMo-RL agents working in /home/ubuntu/RL with limited workspace disk, a larger /ephemeral volume, and optional /home/ubuntu/RL/.env secrets. Use when running nemo-rl-auto-research campaigns, experiments, training jobs, model or dataset downloads, shared cache-heavy commands, log-producing runs, checkpoint generation, W&B or Hugging Face authenticated workflows, or any workflow that may create large files on Brev.
nemo-rl-session-memory
Manage durable working-session memory for coding agents. Use when a user asks to preserve or recover agent context across disconnects, VS Code restarts, long-running work, handoffs, or any session where important state should be written periodically under the repo's session directory. Do NOT use for: simple questions, short tasks, one-off commands, linting, or code review.
tilegym-adding-cutile-kernel
Add a new cuTile GPU kernel operator to TileGym. Covers dispatch registration in ops.py, cuTile backend implementation, __init__.py exports, test creation, and benchmark in tests/benchmark. Use when adding, creating, or implementing a new cuTile operator/kernel in TileGym, or when asking how to register a new cuTile op.
tilegym-converting-cutile-to-julia
Converts cuTile Python GPU kernels (@ct.kernel) to cuTile.jl Julia equivalents. Handles kernel syntax translation, 0-indexed to 1-indexed conversion, broadcasting differences, memory layout (row-major to column-major), type system mapping, and launch API differences. Use when converting, porting, or translating cuTile Python kernels to Julia cuTile.jl, or debugging/optimizing existing Julia cuTile translations.
tilegym-cutile-autotuning
Use when adding, modifying, optimizing, or debugging CuTile autotuning code. Trigger signals: `exhaustive_search` / `replace_hints` / `hints_fn` / `cuda.tile.tune` in code, `autotune` in filenames, or correctness/performance issues in autotuned CuTile kernels. Covers: tune-once/cache/launch pattern, per-architecture configs (sm80–sm120), parameter space design (tile sizes, occupancy, num_ctas), and 7 common pitfalls with solutions.
nemo-rl-docs
Documentation conventions for NeMo-RL. Covers docs/index.md updates and docstring format. Do NOT use for: bug fixes, test fixes, dependency bumps, refactoring, CI/CD changes, performance tuning, or any task that does not involve writing or updating documentation.
agent-teams
创建和管理 Agent Teams 并行任务团队。当用户说"建个团队"、"拉个队伍"、"并行跑"、"create team"、"spawn agents"、"组队"时触发。
bot-config
管理 CloseCrab bot 配置（Firestore）。添加新 bot、给已有 bot 添加 channel、切换 channel、查看配置。当用户说"添加一个 bot"、"新建 bot"、"加个飞书 channel"、"加个 discord channel"、"切换到飞书"、"bot 配置"、"add bot"、"add channel"等关键词时触发。
browser-cli
Drive the already-logged-in Chrome on Chris's cloudtop from the command line (agent-browser over CDP), instead of chrome-devtools-mcp. Use for ANY interactive browser work on internal/SSO-gated sites — gHire, Buganizer UI, Google Chat, internal dashboards, form filling, clicking through a flow. ~40-50x cheaper than the MCP (350 tokens vs 15-20K per snapshot). Also provides multi-round Google Chat conversation — send a message and block waiting for the peer's reply. Trigger on "操作浏览器", "点开网页", "帮我填一下", "去 gHire", "给 X 发个 Chat 消息", "等他回复". Keep chrome-devtools-mcp only for Lighthouse / heap snapshot / performance trace.
chinese-dictation
Generate Chinese (traditional/simplified) dictation practice — extract words+pinyin from a screenshot, produce an HTML page with audio that reads each word aloud with configurable pause between words. Use when user says "听写"、"聽寫"、"语文听写"、"听写练习"、"dictation"、"做个听写"、"听写卡"、"听写音频".
code-wiki-recon
First-pass codebase reconnaissance using Google Code Wiki (codewiki.google). Use when the user is opening a new repository for the first time, asks for a quick architectural overview of an unfamiliar codebase, or needs to build a mental model before diving into code. Pulls the auto-generated TOC, architecture diagram, and section summaries from Code Wiki to bootstrap understanding. Strictly a recon tool, not a Q&A oracle — Code Wiki cannot answer deep runtime/optimization questions, only describe architecture.
deck-builder
Build polished, on-brand slide decks (python-pptx → Google Slides) and richly formatted Google Docs, with a reusable design system, matplotlib charts, AI-generated imagery (Nano Banana), a render-verify loop, and link-stable Drive uploads (PATCH same fileId so the share link never changes). Use when the user wants to create or iterate a presentation / briefing deck / exec one-pager / formatted Google Doc — especially customer or executive briefings that go into Google Drive. Also covers extracting user-pasted images from the session transcript and converting decks/docs to PNG for visual self-check.
cc-pages-backup
CC Pages 增量备份到 GitHub private repo
bwrap-bypass
绕过 Claude Code bwrap (bubblewrap) sandbox 限制。bwrap sandbox 会重新挂载文件系统，导致 Bash 命令无法访问特定路径、凭证文件、Unix socket 等资源。当遇到以下情况时使用：(1) Bash 命令报 Permission denied 但 sandbox 外正常 (2) LOAS2/Kerberos 凭证不可用 (3) 新机器/Bot 部署后需要配置 sandbox bypass (4) 命令因 sandbox 文件系统隔离而失败 (5) 用户说"装 bwrap bypass"、"配 sandbox"。关键词："bwrap"、"sandbox"、"Operation not permitted"、"Permission denied"、"credentials-cache"。
chat-style
聊天平台消息格式化规则。当 Claude Code 通过聊天平台（Discord、飞书等）运行时自动适配消息风格，避免渲染问题。当用户说"聊天风格"、"chat风格"、"消息格式"时触发。
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.