122 件 ・ 関連度順
概要と使いどころ
Manages context passing between DAG nodes and spawned agents. Handles context summarization, selective forwarding, and token budget optimization. Activate on 'bridge context', 'pass context', 'summarize context', 'context management', 'agent context'. NOT for execution (use dag-parallel-executor) or aggregation (use dag-result-aggregator).
日本語の概要は準備中です。原文の説明を表示しています。
Trains large language models (2B-462B parameters) using NVIDIA Megatron-Core with advanced parallelism strategies. Use when training models >1B parameters, need maximum GPU efficiency (47% MFU on H100), or require tensor/pipeline/sequence/context/expert parallelism. Production-ready framework used for Nemotron, LLaMA, DeepSeek.
日本語の概要は準備中です。原文の説明を表示しています。
Trains large language models (2B-462B parameters) using NVIDIA Megatron-Core with advanced parallelism strategies. Use when training models >1B parameters, need maximum GPU efficiency (47% MFU on H100), or require tensor/pipeline/sequence/context/expert parallelism. Production-ready framework used for Nemotron, LLaMA, DeepSeek.
日本語の概要は準備中です。原文の説明を表示しています。
親イシュー配下のサブイシュー（孫含む）を依存順を保ちつつ worktree で並列に自動実装・push 前 review・PR 作成・CI 監視・マージ可能状態化まで一括自動化。 「イシューツリーを並列実装」「配下のサブイシューをまとめて実装」「ツリー全体を並列で実装して」「イシュー階層を自動開発」で使用。 per-issue 計画立案（Plan: セッション継承モデル）→実装（Implement: sonnet）の分業。push 前 review（Review 通過後にのみ push・PR 作成して CI を 1 回だけ起動）。 外部チェック構成は args の externalChecks で明示（{"app", "context"} の組で宣言。[] で「なし」を確定して不要待機なし・未指定なら自動マージ停止・slug のみの旧形式は自動マージ fail-closed）。 自動 squash merge は autoMerge: true + externalChecks 明示（全 App の信頼済み context 宣言込み）の opt-in で実行（merge-exec の自己取得再検証 + サーバー側 branch protection 実測が前提。既定 false はマージ可能状態で停止し人間がマージ。サーバー側 workflow サンプル（upstream の docs/implement-issue-tree/auto-merge-sample.yml 参照）+ branch protection への委譲も可）。並列度（parallel）と依存（dependsOn）で実行順を制御。 Phase 単位で厳密に直列化したい場合は phaseGate: true（opt-in。ルート直下の子〔Phase 親〕を sub-issues リスト順に直列化し、前 Phase の全子孫が merged/closed になるまで次 Phase 配下に着手しない。既定 false は現行の post-order 優先度のみ）。 単一イシューの実装は implement-issue、PR レビューは implement-review-pr を参照。
大きな機能開発や移行の目標を、複数回の作業に分けた実行計画にします。各工程の背景、依存関係、検証方法をまとめ、別のエージェントへの引き継ぎを支えます。
Provides PyTorch-native distributed LLM pretraining using torchtitan with 4D parallelism (FSDP2, TP, PP, CP). Use when pretraining Llama 3.1, DeepSeek V3, or custom models at scale from 8 to 512+ GPUs with Float8, torch.compile, and distributed checkpointing.
日本語の概要は準備中です。原文の説明を表示しています。
Provides PyTorch-native distributed LLM pretraining using torchtitan with 4D parallelism (FSDP2, TP, PP, CP). Use when pretraining Llama 3.1, DeepSeek V3, or custom models at scale from 8 to 512+ GPUs with Float8, torch.compile, and distributed checkpointing.
日本語の概要は準備中です。原文の説明を表示しています。
Decomposes a contextualized plan into atomic, independently executable tasks with complete embedded context, registered through the plan API. Use after SAM Stage 3 Context Integration produces the contextualized plan artifact — when the plan is ready for task generation with CLEAR ordering, CoVe checks, and dependency graphs for parallel execution.
日本語の概要は準備中です。原文の説明を表示しています。
Guide for optimizing MSBuild build parallelism and multi-project scheduling. Only activate in MSBuild/.NET build context. USE FOR: builds not utilizing all CPU cores, speeding up multi-project solutions, evaluating graph build mode (/graph), build time not improving with -m flag, understanding project dependency topology. Note: /maxcpucount default is 1 (sequential) — always use -m for parallel builds. Covers /maxcpucount, graph build for better scheduling and isolation, BuildInParallel on MSBuild task, reducing unnecessary ProjectReferences, solution filters (.slnf) for building subsets. DO NOT USE FOR: single-project builds, incremental build issues (use incremental-build), compilation slowness within a project (use build-perf-diagnostics), non-MSBuild build systems. INVOKES: binlog MCP server tools (expensive_projects, expensive_targets, project_target_times); falls back to dotnet build -m, dotnet build /graph, binlog replay + grep.
日本語の概要は準備中です。原文の説明を表示しています。
Provides PyTorch-native distributed LLM pretraining using torchtitan with 4D parallelism (FSDP2, TP, PP, CP). Use when pretraining Llama 3.1, DeepSeek V3, or custom models at scale from 8 to 512+ GPUs with Float8, torch.compile, and distributed checkpointing.
日本語の概要は準備中です。原文の説明を表示しています。
Manage context-mode GitHub issues, PRs, releases, and marketing with parallel subagent army. Orchestrates 10-20 dynamic agents per task. Use when triaging issues, reviewing PRs, releasing versions, writing LinkedIn posts, announcing releases, fixing bugs, merging contributions, validating ENV vars, testing adapters, or syncing branches.
日本語の概要は準備中です。原文の説明を表示しています。
This skill should be used when writing, enhancing, or evaluating the launch prompt for a long-running autonomous agent or a parallel multi-agent orchestration attacking a hard problem: pseudo-formal task briefs that define terms and an exact success predicate linguistically, enumerate non-counting outcomes, set persistence rules with explicit stop and return conditions and effort floors, manage a diverse portfolio of parallel approaches with an approach registry and blocked-route bookkeeping, and gate the return on adversarial audit. Route agent topology and coordination protocols to multi-agent-patterns, runtime control surfaces and loop governance to harness-engineering, evaluator and quality-gate construction to evaluation, judge design to advanced-evaluation, and compaction or memory mechanics to context-compression and memory-systems.
日本語の概要は準備中です。原文の説明を表示しています。
Operational guide for enabling hierarchical context parallelism in Megatron-Bridge, including config knobs, code anchors, pitfalls, and verification.
日本語の概要は準備中です。原文の説明を表示しています。
Queue processing with fresh context per phase. Processes N tasks from the queue, spawning isolated subagents to prevent context contamination. Supports serial, parallel, batch filter, and dry run modes. Triggers on "/ralph", "/ralph N", "process queue", "run pipeline tasks".
日本語の概要は準備中です。原文の説明を表示しています。
Parallel multi-AI cross-validation research workflow (大版本). Dispatch N internal sub-agents + grok + gemini in parallel, automatically cross-validate findings, tier by confidence (strong consensus / partial / conflict / insufficient), generate tiered action items with arbitration. Use when user says "多 AI 调研", "交叉验证", "独立共识", "三脑调研", "multi-ai research", "parallel research", "cross-validate", or needs deep research that benefits from internal data + external 2026 consensus. NOT for quick factual Q&A, pure code reasoning, or tasks needing deep project context.
日本語の概要は準備中です。原文の説明を表示しています。
Chooses how to launch subagents from Claude Code (the Agent tool, a Workflow script, or headless claude -p children) and runs a headless fan-out from one background command. Use for "fan this out to 50 agents", "run these tasks in parallel", "launch subagents for each file", "should I spawn agents or write a workflow", "batch this through Haiku", "run claude -p in parallel", "dispatch a task list without the launches eating my context", or before any launch of more than a handful of subagents, where the launching itself is the cost. Covers the parent-turn cost of each mechanism, concurrency ceilings, what context and tools each child gets, effort control, output contracts, a cost ledger, and resume after interruption.
日本語の概要は準備中です。原文の説明を表示しています。
Skill-aware orchestration with context routing. Decomposes complex tasks into skill-typed subtasks, extracts targeted context subsets, executes subagents in parallel, and synthesizes results. Self-answers trivial lookups inline. No SDK dependency — uses raw HTTP via httpx. Use when tasks require multiple analytical perspectives, when context is large and subtasks only need portions, or when orchestrating-chat-agents spawns too many redundant subagents.
日本語の概要は準備中です。原文の説明を表示しています。
Use when the user wants a cited, structured read of local documents and project knowledge. Triggers on: "analyze these docs", "scan my project for context", "read the docs folder", "summarize what's in .beagle/concepts/", "extract context from docs/", "what's in this folder", "go read everything in X and tell me what's there". Also invoked programmatically by other beagle skills (prfaq-beagle Ignition, brainstorm-beagle reference points, strategy-interview context grounding) via the companion contract. Does NOT trigger on codebase lookups ("find this function", "search the repo"), web research (use web-research), LLM-as-judge evaluation (use llm-judge), or document editing (use humanize-beagle). Produces a written scan plan, parallel-subagent findings, and a cited synthesis report on disk — never inline prose, never unsourced claims.
日本語の概要は準備中です。原文の説明を表示しています。
Fornece treinamento distribuído de LLM nativo do PyTorch usando torchtitan com paralelismo 4D (FSDP2, TP, PP, CP). Use para fazer pré-treinamento de Llama 3.1, DeepSeek V3 ou modelos customizados em escala de 8 a 512+ GPUs com Float8, torch.compile e checkpoint distribuído.
日本語の概要は準備中です。原文の説明を表示しています。
Treina modelos de linguagem grandes (2B-462B de parâmetros) usando NVIDIA Megatron-Core com estratégias avançadas de paralelismo. Use ao treinar modelos >1B de parâmetros, precisar de máxima eficiência de GPU (47% MFU em H100), ou necessitar de paralelismo tensor/pipeline/sequência/contexto/expert. Framework pronto para produção usado em Nemotron, LLaMA, DeepSeek.
日本語の概要は準備中です。原文の説明を表示しています。
Upstash サーバーレスデータプラットフォームリファレンス。 @upstash/redis — REST API、pipelining、transactions、JSON、Search、グローバルレプリケーション。 @upstash/ratelimit — Fixed Window、Sliding Window、Token Bucket、limit、blockUntilReady。 QStash — publishJSON、schedules、queues、DLQ、URL Groups、callbacks、flow-control。 @upstash/vector — upsert、query、ANN、hybrid index、sparse index、embedding models、namespace。 Rust 製ローカル vector DB エンジン fandhe-db とは別（こちらはマネージド SaaS の JS SDK）。 @upstash/workflow — durable execution、serve、context.run/sleep/call/invoke、waitForEvent、parallel steps、agents。 Upstash Search — search、upsert、fetch、range、filtering、reranking、algorithm、@upstash/search。 Upstash Box — サンドボックス、agent、filesystem、git、browser（CDP/AI actions/recordings）、network policy、snapshots、schedules。
Bulk mode — maximum development speed by spending tokens aggressively on massive parallel sub-agents and multi-agent workflows, while the orchestrator context stays lean (pairs with /slim). Invoking /bulk is the user's explicit opt-in to Workflow/ultracode-style orchestration for this session. Use when the user types /bulk, asks for "爆速", "並列で大量に", "ultracode", "多エージェント", or hands over a big task where wall-clock speed matters more than token cost.
Full PR lifecycle in a fresh task-owned git worktree: implement via the ulw-loop skill with mandatory evidence-bound manual QA → reviewer-readable English PR → verification loop (CI + Cubic, where Cubic is skipped only when its quota is exhausted) → merge by default → worktree cleanup. Decomposes one task into the smallest atomic, independently-mergeable PRs and builds the independent ones concurrently via one worktree per PR driven by parallel subagents or a team. Unbounded loop: any failing gate sends you back to fix-and-re-QA inside that PR's worktree. Use whenever implementation work needs to land as a PR. Triggers: 'create a PR', 'implement and PR', 'work on this and make a PR', 'implement issue', 'land this as a PR', 'split into atomic PRs', 'parallel PRs', 'work-with-pr', 'PR workflow', 'implement end to end', even when user just says 'implement X' if the context implies PR delivery.
日本語の概要は準備中です。原文の説明を表示しています。
RNN+Transformer hybrid with O(n) inference. Linear time, infinite context, no KV cache. Train like GPT (parallel), infer like RNN (sequential). Linux Foundation AI project. Production at Windows, Office, NeMo. RWKV-7 (March 2025). Models up to 14B parameters.
日本語の概要は準備中です。原文の説明を表示しています。