1.4K skills · sorted by relevance
Coordinate multiple coding agents through Superset in separate workspaces. Split work into bounded tasks, track dependencies and progress, send follow-ups, and collect structured results covering changed files, verification checks, blockers, and the order for combining work.
Implement and troubleshoot large-model training with DeepSpeed across multiple GPUs. The skill covers memory optimization, parallel training, and numerical precision choices, alongside examples for storage transfers and models that divide work among specialized components.
Split independent research or data-entry tasks among parallel workers. Provides a workflow for tracking completion in a table, testing one task first, and using a Python playbook to retry unfinished work and resume interrupted runs.
Run independent development tasks in parallel, each with a separate git worktree and implementation agent. Each agent opens its own pull request, which receives cross-review before being handed to a human for review and merging.
Coordinate temporary agents for code exploration, independent implementation tasks, verification, reviews, or complex debugging. Define self-contained assignments, run independent work in parallel, and check returned evidence before combining the results.
Search the web and specialized subject areas, run multiple searches in parallel, and extract page text from URLs. Use the bundled command-line tools to gather current information, check claims, and read beyond search snippets.
Carry out a validated development plan in an isolated Git worktree, a separate working directory. Coordinate parallel implementation, run quality checks, update documentation, and take the changes through pull request creation, merging, execution metrics, and cleanup.
Codex-only team orchestration: run a named team of cooperating Codex workers with durable, script-managed state. MUST USE when the user asks Codex to create, run, coordinate, inspect, archive, or delete a team of agents/threads/sessions, or to work on something as a team in parallel. FIRST inspects the active tool surface (checking tool_search for deferred tools) and tells the user the route: native MultiAgentV2 agents (flat spawn_agent with task_name) when available, Codex App threads as the fallback, or a plain-subagent split when neither set exists. The main session is always the leader; members are defined by a concrete part, ownership area, or perspective - never a vague job role; a bundled cross-platform script writes the .omo/teams state plus an auto-generated member field manual. Use a team when the work is not perfectly isolated but parallelizing helps; use plain subagents when scope is perfectly isolated or the goal is ambiguous. Triggers: team mode, teammode, make a team, run as a team, team of agents, coordinate threads, parallel Codex threads, archive the team.
The English summary is being prepared. Showing the original description.
Parallel/distributed computing. Scale pandas/NumPy beyond memory, parallel DataFrames/Arrays, multi-file processing, task graphs, for larger-than-RAM datasets and parallel workflows.
The English summary is being prepared. Showing the original description.
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.
The English summary is being prepared. Showing the original description.
Operational guide for choosing and combining parallelism strategies in Megatron Bridge, including sizing rules, hardware topology mapping, and combined parallelism configuration.
The English summary is being prepared. Showing the original description.
Coordinates parallel investigation threads to simultaneously explore multiple hypotheses or root causes across different system areas. Use when debugging production incidents, slow API performance, multi-system integration failures, or complex bugs where the root cause is unclear and multiple plausible theories exist; when serial troubleshooting is too slow; or when multiple investigators can divide root-cause analysis work. Provides structured phases for problem decomposition, thread assignment, sync points with Continue/Pivot/Converge decisions, and final report synthesis.
The English summary is being prepared. Showing the original description.
Patterns for parallel subagent execution using the Agent tool (formerly Task). Use when coordinating multiple independent tasks, spawning dynamic subagents, or implementing features that can be parallelized.
The English summary is being prepared. Showing the original description.
Use this skill to run multiple Simulink simulations in parallel with parsim, batchsim, and DesignStudy. TRIGGER when: user asks for parameter sweep, sweeping/varying parameters across simulations, Monte Carlo study, batch simulation, running multiple simulations, parallel simulations, large-scale multiple simulations, any task involving repeated simulations with different parameter values, user mentions having an existing script/function for parameter sweeps, or user wants to migrate from parfor/set_param/for-loop patterns to recommended workflows.
The English summary is being prepared. Showing the original description.
Speed up local parfor, parfeval, or spmd by switching to a thread-based parallel pool. Trigger when a user describes slow or disappointing local parallel performance, even if they don't mention threads. Symptoms: parfor on a laptop/workstation is slower than expected or "only slightly faster than for"; parfor scales poorly with the number of workers; ticBytes/tocBytes, the Parallel Pool dashboard, mpiprofile, or system tools show large per-worker data transfer; large broadcast variables or sliced inputs make iterations slow; opening a process pool dominates a short workload; user mentions serialisation or data transfer overhead. Also trigger on any question about whether code or a function works on a thread pool. For non-pool MATLAB performance work (vectorisation, preallocation, profiling), defer to matlab-optimize-performance.
The English summary is being prepared. Showing the original description.
Decision protocol for the map-reduce / dynamic fan-out pattern in LM pipelines — "given list L, run f(item) for each item in parallel, then combine". Activates when the coder agent is about to process N items with N LM calls (per-doc summarize, per-query retrieve, per-candidate rank, parallel tool fan-out). Encodes the *when*, *how many at once*, *what to do when one fails*, and *how to reduce* — not the API of any single framework. Cross-framework: LangGraph `Send`, CrewAI parallel tasks / Flow, `asyncio.gather`, `ThreadPoolExecutor`, LlamaIndex batch retrieval.
The English summary is being prepared. Showing the original description.
Tracks per-agent token usage and flags waste in parallel dispatch. Use when evaluating parallel agent efficiency or after a multi-agent run.
The English summary is being prepared. Showing the original description.
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.
The English summary is being prepared. Showing the original description.
Plan, implement, or review dispatching parallel agents work in an existing codebase with compatibility, security, and verification controls. Use when the user explicitly requests dispatching parallel agents work.
The English summary is being prepared. Showing the original description.
Guides parallel exploration of multiple implementation approaches using git worktrees. Use when facing decisions with multiple valid paths. Triggers on: explore options, compare approaches, parallel exploration.
The English summary is being prepared. Showing the original description.
OPTIONAL per-task isolation for autonomous parallel work. Routed to only on explicit opt-in by subagent-driven-development or dispatching-parallel-agents, so parallel units mutate files without colliding. Stands up one worktree per unit, works it in isolation, then integrates each unit back onto the caller's working branch for the caller's single commit-gate + finishing-a-development-branch exit. Never on the default path; it does not bypass a gate or finish per unit.
The English summary is being prepared. Showing the original description.
The parallel fan-out primitive. Routed to by any skill or command that splits work across independent units and dispatches an agent per unit — subagent-driven-development, /sprint, parallel /review. It owns the dispatch/collect/funnel discipline: bound concurrency, isolate units, collect every result, dedupe overlap, and funnel through finding-triage then the read-only verdict-aggregator. Raw agent output is never consumed before the funnel runs; an agent that errors records its unit as incomplete without corrupting the batch.
The English summary is being prepared. Showing the original description.
Splits a coding task into independent parallel agent workstreams with disjoint write sets, explicit inputs and outputs, dependency ordering, and a merge/review plan. Use when several subtasks can proceed without shared mutable state. Not for tightly coupled edits, one failing root cause, or parallel changes to the same files.
The English summary is being prepared. Showing the original description.
Decompose a task into safe parallel sub-agent slices with explicit ownership, optional worktree isolation, per-worker plans, verification, and orchestrator-led integration. Use when the user explicitly asks for delegation, sub-agents, or parallel work and the task can be split into non-overlapping scopes.
The English summary is being prepared. Showing the original description.