nccl_culprit_victim

nccl_culprit_victim is a skill for Claude Code, Codex from DeepLink-org/probing. It costs 27 tokens per session (481 once invoked), scanned A, original, Apache-2.0.

A distributed-training diagnostic that uses NCCL profiler data to trace communication waits back to a slow sender or a process waiting on others. NCCL is the library commonly used for GPU-to-GPU communication.

In plain words
What is it for?
Use it to inspect send and receive waits, peer relationships, propagated delays, and possible receiver congestion in recent collective operations.
Why use it?
It distinguishes the process causing the delay from one merely experiencing its effects, which narrows the cause of a communication slowdown.

Skill for Claude CodeCodex

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

Good fit Use it to inspect send and receive waits, peer relationships, propagated delays, and possible receiver congestion in recent collective operations.

Compare 6 skills from other repositories ↓
Install with agentmods
npx agentmods add skills/deeplink-org/probing/nccl_culprit_victim
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 DeepLink-org/probing --skill nccl_culprit_victim
Clone the repo
git clone --depth 1 https://github.com/DeepLink-org/probing

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 nccl_culprit_victim

README.md
[![agentmods](https://agentmods.dev/badge/skills/deeplink-org/probing/nccl_culprit_victim/github.svg)](https://agentmods.dev/skills/deeplink-org/probing/nccl_culprit_victim)
Your own site
<a href="https://agentmods.dev/skills/deeplink-org/probing/nccl_culprit_victim"><img src="https://agentmods.dev/badge/skills/deeplink-org/probing/nccl_culprit_victim/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 nccl_culprit_victim

Your own site · 80×15
<a href="https://agentmods.dev/skills/deeplink-org/probing/nccl_culprit_victim"><img src="https://agentmods.dev/badge/skills/deeplink-org/probing/nccl_culprit_victim.svg" alt="Reviewed on agentmods" width="80" height="20"></a>
Per session 27 Skills are progressive disclosure: only the name and description are preloaded; the body loads when the skill is used.
When invoked 481 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.
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 $0.00027 $0.00481
Opus 5 $0.00014 $0.00241
Sonnet 5 $0.00005 $0.00096
Haiku 4.5 $0.00003 $0.00048

Measured 11d ago against content hash 42091062862c, method: parsed. Prices are Anthropic first-party input rates as of 2026-09-10, from the pricing page.

Security

Grade A, and why

nccl_culprit_victim 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 11d 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.

python/probing/bundled_skills/nccl_culprit_victim/SKILL.md · 38 lines

What it actually says

NCCL culprit / victim attribution

Uses NCCL profiler plugin wait decomposition (nccl.proxy_ops) with peer-topology attribution (rank↔peer edges, not global wait sorting):

  • Culprit (slow local GPU): high send_gpu_wait_ns on a send edge (rank → peer)
  • Victim (waiting on peers / network): high recv_wait_ns on recv edge; check attributed_root_rank
  • Propagated victim: recv wait high but peer's send_gpu_wait_ns is also high — root cause is upstream
  • Receiver congestion (v4 ABI only): high send_peer_wait_ns — this rank's sends are stalled waiting for the peer's clear-to-send credits; investigate that peer

Requires NCCL_PROFILER_PLUGIN + probing-nccl-profiler (or mock data via PROBING_NCCL_MOCK on macOS).

Parameters

  • seq_window (integer, default 20): recent collective sequences to analyze
  • use_global (boolean, default true): fan-out with global.nccl.proxy_ops
  • Coarse collective latency → comm_bottleneck / slow_rank (python.comm_collective)
  • IB QP timing → nccl.net_qp when NetPlugin mask enabled
Files

What ships with it

1 file 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. 11d ago First seen · 38 lines · 27 tokens per session scan A 42091062862c

Subscribe to this mod's changes

nccl_culprit_victim is a skill published in the GitHub repository DeepLink-org/probing (11 stars, last pushed 5d ago), licensed Apache-2.0. It adds 27 tokens to every session and 481 once invoked, about $0.0001 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 2026-08-30.

Related

Other skills, from other repositories

mcp-apps-builder

MANDATORY for ALL MCP server work - mcp-use framework best practices and patterns. READ THIS FIRST before any MCP server work, including: Creating new MCP servers Modifying existing MCP servers (adding/updating tools, resources, prompts, widgets) Debugging MCP server issues or errors Reviewing MCP server code for…

Shubhamsaboo/awesome-llm-apps · 139 tokens

developing-with-streamlit

Use for ALL Streamlit tasks: creating, editing, debugging, beautifying, styling, theming, optimizing, or deploying Streamlit apps. Also custom components, st.components.v2, HTML/JS/CSS work. Discovers and loads version-matched reference docs from the user's installed Streamlit (>=1.57). Triggers: streamlit, st.…

streamlit/streamlit · 128 tokens

python-code-quality

Code quality checks, linting, formatting, and type checking commands for the Agent Framework Python codebase. Use this when running checks, fixing lint errors, or troubleshooting CI failures.

microsoft/agent-framework · 40 tokens

eve

Build durable backend AI agents with the eve framework. Use when creating, editing, or debugging an eve project — agent instructions, skills, tools, connections, channels, sandboxes, subagents, schedules, or evals.

ComposioHQ/composio · 47 tokens

plugin-architecture-patterns

Design, implement, or diagnose Xberg plugin traits, typed registries, priority collisions, lifecycle, native extractors, and Alef-generated Python plugin bridges. Load for plugin-system work, not ordinary extractor parsing.

xberg-io/xberg · 49 tokens

test-corpus

The testdocuments submodule is a bucket-fetched fixture corpus that is not committed. This skill covers readtestfixture, missing fixtures, valid A/B controls, and submodule push order. Load before running Rust tests on a fresh clone, setting up an A/B control, adding a fixture-backed test, or diagnosing…

xberg-io/xberg · 72 tokens