Browse reusable Agent Skills, each with a clear purpose and practical guidance.
Adds sharding-aware IR hints (op substitutions, sharding kwargs, all_reduce insertions) directly into an existing AutoDeploy custom model (modeling_*.py). Edits the file in place — no separate _ir.py copy. Validates with apply_sharding_hints and end-to-end multi-GPU runs.
Onboard a HuggingFace multimodal model (vision/audio/video + text) to the TensorRT-LLM PyTorch backend. Use when writing a new `tensorrt_llm/_torch/models/modeling_<vlm>.py` plus its input processor and weight mapper, or extending an existing VLM. Not for AutoDeploy — use `ad-model-onboard` for that path.
Review, design, and refactor TensorRT-LLM PyTorch MoE code for architecture fit, clean code, maintainability, and testability. Always use for any modification, review, refactor, or design planning that touches MoE modules, including tensorrt_llm/_torch/modules/fused_moe, ConfigurableMoE, MoE backends, MoEScheduler/moe_scheduler.py, forward execution/chunking, communication strategies, EPLB, quantization/weight handling, routing, factories, MoE docs, or MoE tests. Also use when the user asks whether a MoE design follows the current architecture or whether a MoE refactor is reasonable.
Evaluate a change to the LLM extraction pipeline against a clean baseline. Use when the user wants to test, validate, or assess a prompt/chunking/post-processing/provider change before committing. Produces a merge / regression / defer verdict.
Evaluate a patched mandoc binary's markdown rendering against the current tools/mandoc-md. Use when the user wants to test, validate, or assess a candidate mandoc build before promoting it. Produces a merge / regression / defer verdict and, on regression, a handoff prompt for an agent in the mandoc source worktree.
Drive a mandoc rendering-bug fix end-to-end — scope the bug class, dispatch a subagent in the mandoc source tree, validate via /eval-render audit + compare, and decide whether to promote. Use when the user has identified (or suspects) a class of `-T markdown` rendering bug and wants the full diagnose → fix → validate → promote cycle.
Draft a design/implementation plan for a change, pressure-test it with /dual-review (design mode), iterate with the user, then commit it under plans/. Use when the user wants to plan something before building it - "plan X", "let's plan the Y refactor", "/my-plan", "work up a plan for Z and review it". This is the planning dance for this repo - it produces a tracked plan doc, not code.
Execute PHP code on 3v4l.org to test runtime behavior across multiple PHP versions. Use when you need to verify what PHP actually does with a piece of code.
Give this agent persistent memory via memU. Use to memorize (save files/folders so context survives this session) when the user asks to remember or sync something, or after completing work worth keeping; and to retrieve (search memory from earlier sessions) when the user asks what is known about a person/project/topic or references context not in this conversation.
Upgrade oxc, run codegen, and fix any breaking changes.