NVIDIA/TensorRT-Model-Connect is a collection of C++ reference implementations for deploying supported Hugging Face models with NVIDIA TensorRT, a system that runs trained models to produce inference results. It is for developers who want to build and run supported models or evaluate model integrations through TensorRT. The catalogue entries provide skills and instructions for using this model deployment workflow.
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 skills add NVIDIA/TensorRT-Model-Connect --skill pr-babysittergit clone --depth 1 https://github.com/NVIDIA/TensorRT-Model-ConnectWrote 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/nvidia/tensorrt-model-connect/pr-babysitter)<a href="https://agentmods.dev/skills/nvidia/tensorrt-model-connect/pr-babysitter"><img src="https://agentmods.dev/badge/skills/nvidia/tensorrt-model-connect/pr-babysitter/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.
<a href="https://agentmods.dev/skills/nvidia/tensorrt-model-connect/pr-babysitter"><img src="https://agentmods.dev/badge/skills/nvidia/tensorrt-model-connect/pr-babysitter.svg" alt="Reviewed on agentmods" width="80" height="20"></a>- NVIDIA SkillSpector warn
SkillSpector: 1 finding, up to high
These are SkillSpector’s own severities. On a checked sample its high-severity flags on skills were ~96% false positives — a documented command, a public API, a “never do X” rule — so we show them as a caution to read, not a verdict. Why →
- high Tool Misuse · line 233 Tool parameters are crafted to achieve unintended or unsafe behavior. Parameter abuse can bypass intended safety checks (e.g. shell=True, --force, dangerous glob patterns).Fix: Validate all tool parameters against an allowlist. Reject dangerous parameter values (shell=True, --force, -rf /) and use safe defaults.
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 | $0.00048 | $0.02861 |
| Opus 5.5 | $0.00019 | $0.01144 |
| Sonnet 5.5 | $0.00010 | $0.00572 |
| Haiku 4.5 | $0.00005 | $0.00286 |
Grade A, and why
pr-babysitter 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 1mo 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 — 318 lines — stays where its author put it; the contents beside it link to each section on GitHub.
PR Babysitter
Purpose
Monitor open non-draft GitHub PRs, classify their CI state, rebase when they are
behind main, diagnose failed checks, make minimal fixes on the PR source
branch, and push updated commits. Work sequentially and report every PR state.
Ground Rules
- Use the
githubremote and targetmain. - Never push to
main. - Never change unrelated PR branches.
- Do not rewrite a PR's intent during rebase. Preserve the feature and adapt it
to current
main. - Do not skip CI unless the user explicitly asks.
- Never request, enable, or queue GitHub auto-merge. Do not use
gh pr merge --autoor any equivalent API flag. - When merge authority is granted, merge only after the latest CI for the current PR head has completed successfully and the PR is mergeable.
- Treat GitHub branch protection and rulesets as insufficient evidence of CI safety. The repository may allow a squash or rebase merge while checks are still queued, pending, running, or absent; that is still forbidden here.
- Use
$write-git-messagesfor fix commits, PR comments, and squash or rebase text. - If a fix needs unavailable hardware, product judgment, or broad scope, stop and report the blocker.
Current CI Contract
Read .github/workflows/internal-ci-bridge.yml from current github/main.
Source keeps only Internal CI Bridge and Pages workflows. Premerge, including
legal compliance, and nightly execution stay in private Internal CI.
Before triggering premerge, capture the current PR metadata head and inspect its existing protected status:
REPOSITORY=NVIDIA/TensorRT-Model-Connect
PR_NUMBER=<number>
pull=$(gh api "repos/$REPOSITORY/pulls/$PR_NUMBER")
PR_HEAD_SHA=$(jq -er '.head.sha' <<<"$pull")
gh api \
"repos/$REPOSITORY/commits/$PR_HEAD_SHA/status" \
--jq '[.statuses[] |
select(.context == "TRTMC Internal CI / Automated premerge gate")][0]'
gh pr edit "$PR_NUMBER" \
--repo "$REPOSITORY" \
--add-label run-internal-ci
If a label event was created for an older SHA, let the bridge report the
superseded trigger, wait for Community CPU / Required on the current PR head,
and retry.
When authorization fails and leaves run-internal-ci attached, remove it before
adding it again; adding an existing label does not emit another label event.
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.
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.
- 1mo ago Changed · +2 lines 214869f8b369
- 1mo ago First seen · 316 lines · 48 tokens per session scan A 270289f1c39e
pr-babysitter is a skill published in the GitHub repository NVIDIA/TensorRT-Model-Connect (276 stars, last pushed today), licensed Apache-2.0. It adds 48 tokens to every session and 2,861 once invoked, about $0.0002 per session on Opus 5.5. 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
kn-doc
Use when working with Knowns documentation - viewing, searching, creating, or updating docs.
huggingface-transformers
Use when running or fine-tuning Hugging Face Transformers models in Python — pipeline() inference for text classification and generation, loading weights with AutoModel/AutoTokenizer, devicemap and dtype memory management, tokenizer usage, Trainer fine-tuning, and BGE embeddings for Chinese RAG. Covers transformers…
ia-c-systems
C patterns for systems code, libraries, and native extensions: module layout, function decomposition, status-enum errors, memory safety, undefined behavior, and performance measurement. Use when writing, reviewing, refactoring, or debugging C, working with malloc lifetimes, buffer overflows, sanitizers, or Valgrind…
lsp-investigate
Investigate code using LSP tools. Register a project, then use definitions, references, call hierarchies, type hierarchies, hover, and diagnostics to understand code structure, trace bugs, and assess change impact.
ia-cpp-systems
Modern C++ patterns: RAII and ownership, rule of zero/five, exceptions and error handling, API and ABI boundaries, templates, and CMake tooling. Use when writing, reviewing, refactoring, or debugging C++, working with smart pointers, move semantics, memory leaks, template errors, or gtest. For plain C, see…
graphsignal-profiler
Set up the Graphsignal Profiler for inference workloads — vLLM, SGLang, PyTorch, and dstack services. Use when the user wants GPU profiling, tracing, or monitoring for inference, asks about graphsignal-run or graphsignal.watch(), or asks about CUPTI / Prometheus / OTLP setup.