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/nvidia/tensorrt-llm/perf-host-optimizationnpx skills add NVIDIA/TensorRT-LLM --skill perf-host-optimizationgit clone --depth 1 https://github.com/NVIDIA/TensorRT-LLMWhat 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 | $0.00052 | $0.03448 |
| Opus 5 | $0.00026 | $0.01724 |
| Sonnet 5 | $0.00010 | $0.00690 |
| Haiku 4.5 | $0.00005 | $0.00345 |
Grade A, and why
perf-host-optimization 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 yesterday.
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.
The source is not reproduced here
No licence file
A repository with no LICENSE is all rights reserved by default, so the body is not copied here. The metadata, the measurements and the link are.
What ships with it
12 files 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.
- references/communication-patterns.md 6.7 KB
- references/examples.md 5.4 KB
- references/hot-path-files.md 5.3 KB
- references/hotspot-classification.md 15 KB
- references/optimization-patterns.md 1.2 KB
- references/optimization-strategy.md 9.9 KB
- references/patterns/compound-pitfalls.md 7.5 KB
- references/patterns/gpu-graph.md 7.1 KB
- references/patterns/loop-iteration.md 7.1 KB
- references/patterns/python-overhead.md 6.3 KB
- references/patterns/sync-alloc.md 5.2 KB
- references/patterns/system.md 5.5 KB
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.
- yesterday First seen · 298 lines · 52 tokens per session scan A d2e544602484
perf-host-optimization is a skill published in the GitHub repository NVIDIA/TensorRT-LLM (14,505 stars, last pushed 2d ago), with no licence file. It adds 52 tokens to every session and 3,448 once invoked, about $0.0003 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.
Other skills, from other repositories
perf-host-optimization
Profiles and optimizes LLM inference host / CPU overhead (vLLM / SGLang / TRT-LLM serve / lmdeploy / TensorRT-LLM PyExecutor) using lineprofiler (with nsys support planned). Runs iterative profile-analyze-optimize-validate rounds. Use when GPU utilization is low, when perf-host-analysis has concluded "host-bound", or…
add-jit-kernel
Step-by-step tutorial for adding a new lightweight JIT CUDA kernel to sglang's jitkernel module.
sglang-runtime-context
How SGLang's runtime configuration and process-global state are organized (RuntimeContext tiers, publish + namespace config bags, the pristine ServerArgs seed, override entry points, resource/stream/buffer leases, per-forward flags), the CI guardrails that enforce the design, and the idioms for developing and testing…
sglang-diffusion-add-model
Use when adding a new diffusion model or Diffusers pipeline to SGLang.
sglang-diffusion-performance
Use when choosing the fastest SGLang Diffusion flags for a model, GPU, and VRAM budget.
llm-torch-profiler-analysis
Unified LLM torch-profiler triage skill for sglang, vllm, TensorRT-LLM, and TokenSpeed. Use it to inspect an existing trace.json(.gz) or profile directory, or to drive live profiling against a running server when supported and return one three-table report with kernel, overlap-opportunity, and fuse-pattern tables.