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.
npx agentmods add skills/zystem-io/zymtrace-skills/optimize-gpu-workloadsnpx skills add zystem-io/zymtrace-skills --skill optimize-gpu-workloadsgit clone --depth 1 https://github.com/zystem-io/zymtrace-skillsWrote 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.
[](https://agentmods.dev/skills/zystem-io/zymtrace-skills/optimize-gpu-workloads)<a href="https://agentmods.dev/skills/zystem-io/zymtrace-skills/optimize-gpu-workloads"><img src="https://agentmods.dev/badge/skills/zystem-io/zymtrace-skills/optimize-gpu-workloads.svg" alt="Measured on agentmods" height="20"></a>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.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00398 | $0.02431 |
| Opus 5 | $0.00199 | $0.01215 |
| Sonnet 5 | $0.00080 | $0.00486 |
| Haiku 4.5 | $0.00040 | $0.00243 |
Grade A, and why
optimize-gpu-workloads 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 6d 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.
How it starts
The opening of the file, as written. The whole thing — 108 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Optimize GPU Workloads
The MCP fetches the data — GPU metrics, flamegraphs; you analyze. Load-bearing discipline: always pull both the GPU and CPU views with the same filter — half the time a GPU workload's bottleneck is host-side.
The common discipline — data-source policy, pre-flight, rank-first vs. drill-down, scope-to-own-code/ROI, the always-recommend-and-apply-the-fix rule, the output-template skeleton, severity sizing, and security — lives in shared/analysis-conventions.md. Read it. This skill adds the GPU-specific protocol and call-tree rendering on top.
Connection setup lives in configure-zymtrace-mcp; this skill assumes the MCP is connected. CPU-only workload (no GPUs)? Use optimize-cpu-workloads — it skips the GPU view, GPU metrics, and inference-server framing entirely.
Standard starter prompt (for users who don't know what to ask)
"Analyze the GPU flamegraph over the last 1 hour and suggest solutions."
If the user hands you anything close to that (or shorter — "what's slow on my GPU", "investigate my GPU"), interpret it as: scope to the last 1 hour, pull the GPU flamegraph, cross-check the CPU view, follow the template. Variations: "Analyze [vLLM / SGLang / Triton / my training job] over the last [Nh / since deploy]", "Where's the bottleneck right now?", "What's wasting GPU time today?". Default to the last 1 hour and the whole cluster if no workload is named (ask which to narrow if results look noisy).
The cross-view protocol
The MCP pulls the data; you do the analysis and the discipline of asking for both sides. Establish a data path first (pre-flight, in the shared doc).
- Pull the workload's GPU metrics first, for context. GPU utilization, memory used/bandwidth, SM efficiency, Tensor-Core activity, temperature, and PCIe throughput — plus CPU utilization — to establish whether the workload is GPU- or host-bound and which view will be informative. If the workload is an inference server, also pull its framework metrics where collected — vLLM, SGLang, and NVIDIA Dynamo-Triton (queue depth / pending requests, running-vs-waiting batch size, tokens/sec, KV-cache utilization, prefix-cache hit rate, time-to-first-token). These tell you whether the GPU is starved, saturated, or memory-bound before you read a single frame. Carry the numbers into the recap; they frame the flamegraph findings. (Interpretation patterns: Inference-server specifics below.)
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.
- 6d ago First seen · 108 lines · 398 tokens per session scan A 7a4d29ac0396
optimize-gpu-workloads is a skill published in the GitHub repository zystem-io/zymtrace-skills (5 stars, last pushed 22d ago), licensed Apache-2.0. It adds 398 tokens to every session and 2,431 once invoked, about $0.0020 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 2026-08-31.
Other skills, from other repositories
debug-cuda-crash
Tutorial for debugging CUDA crashes using API logging.
test-amplification
Methodical bug-driven test amplification for pygraphistry features. Use when hardening a feature area via user-workflow exploration, 5-Whys retrospectives, bug-taxonomy derivation, concrete test planning, implementation, and safety-gated validation.
pt2-bug-basher
Debug PyTorch 2 compiler stack failures including Dynamo graph breaks, Inductor codegen errors, AOTAutograd crashes, and accuracy mismatches. Use when encountering torch.compile errors, BackendCompilerFailed exceptions, recompilation issues, Triton kernel failures, FX graph problems, or when the user mentions…
aoti-debug
Debug AOTInductor (AOTI) errors and crashes. Use when encountering AOTI segfaults, device mismatch errors, constant loading failures, or runtime errors from aotcompile, aotload, aoticompileandpackage, or aotiloadpackage.
fix-issue
Fix bugs reported in PyTorch GitHub issues by reproducing, root-causing, and implementing a fix in the local working tree. Use when the user asks to fix a PyTorch GitHub issue.
cuda-kernel-optimizer
Use when optimizing, tuning, diagnosing, or profiling CUDA, CUTLASS, Triton, PyTorch, vLLM, TensorRT-LLM, or another GPU workload; when assessing an NCU, Nsys, or PyTorch Profiler report; or when the test workload, correctness checks, measurement path, or target environment is incomplete.