菜单
Orchestrates work on the University of Leeds AIRE HPC cluster using the aire-agent toolkit. Use when starting AIRE sessions, submitting Slurm jobs, choosing between raw Slurm and aire-agent tools, wiring MCP for Claude Code, or following generate-validate-submit-debug loops on AIRE.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | aire-agent-workflow |
| description | Orchestrates work on the University of Leeds AIRE HPC cluster using the aire-agent toolkit. Use when starting AIRE sessions, submitting Slurm jobs, choosing between raw Slurm and aire-agent tools, wiring MCP for Claude Code, or following generate-validate-submit-debug loops on AIRE. |
Act as an AIRE cluster assistant. The install root is ${AIRE_AGENT:-$HOME/.aire-agent} (call it AIRE below).
Start substantial tasks with a short note: task type, partition/GPU needs, and whether the user is on a login node or local machine.
--gres=gpu:N, N≤3); use multi-node for more.--partition=gpu and --gres=gpu:N together on GPU jobs.--time required on every job.srun.$TMP_SHARED is wiped when the job ends — copy artifacts to $SCRATCH.$HOME; datasets/checkpoints on $SCRATCH.Read $AIRE/knowledge/ and $AIRE/AGENTS.md before guessing policy.
| Task | Command |
|---|---|
| Submit | sbatch script.sh |
| Queue | squeue --me |
| Cancel | scancel <job_id> |
| Efficiency | seff <job_id> |
| Quota | quota -s |
| GPU nodes | sinfo -p gpu |
Do not wrap these in custom tools unless the user asks for CLI helpers (aire-agent queue, etc.).
export PATH="$AIRE/bin:$PATH"
aire-agent generate --gpu 1 --time 4h --framework pytorch
aire-agent validate my_job.sh
aire-agent search "GPU memory"
aire-agent log --name run1 --metrics '{"loss":0.3}' --params '{"lr":0.001}'
aire-agent doctor
Register once:
claude mcp add aire "$AIRE/mcp/server.py"
Registered MCP tools (see $AIRE/mcp/server.py):
| Tool | Purpose |
|---|---|
generate_script | SBATCH with AIRE constraints |
validate_script | Pre-submit checks |
search_docs | Knowledge base search |
list_modules | Module list |
system_info | Hardware summary |
log_experiment | JSONL experiment log |
query_experiments | Query past runs |
sync_docs | Refresh upstream docs |
Job queue/submit tools are intentionally not in MCP — use Slurm directly.
aire-conda-environments; optional aire-github-installs.aire-agent generate or MCP generate_script.aire-agent validate or MCP validate_script.sbatch script.sh.squeue --me; tail logs/.seff <job_id>; skill aire-ddp-debugging if distributed.aire-agent log or MCP log_experiment.| User need | Load skill |
|---|---|
| Conda / env YAML | aire-conda-environments |
| GitHub pip / editable install | aire-github-installs |
| Multi-GPU training | aire-l40s-distributed-training |
| DDP hang / NCCL | aire-ddp-debugging |
| Repo layout / Hydra | aire-research-software-engineering |
| Medical CV science | medical-cv-research-engineer (domain, not AIRE ops) |
$AIRE/CLAUDE.md — Claude Code session behavior$AIRE/AGENTS.md — generic agent rules$AIRE/README.md — install and examplesDesigns PyTorch distributed training on AIRE NVIDIA L40S GPUs (Ada, 48GB, PCIe, max 3 GPUs per node). Covers DDP vs FSDP, bf16, batch sizing, NCCL and torchrun with Slurm, TMP_SHARED staging, and SCRATCH checkpointing. Use for multi-GPU or multi-node training on AIRE gpu partition.
Installs Python packages and research code from GitHub on AIRE HPC. Use for pip install git+https, editable installs (-e), git submodules, private repositories, CUDA/C++ extension builds, or import failures between login and compute nodes on AIRE.
Manages Miniforge/conda environments on the University of Leeds AIRE HPC cluster. Use when creating, updating, exporting, or debugging Python/R conda envs on AIRE; wiring activation into SBATCH or srun; fixing HOME quota; or choosing Miniforge, conda-pack, Apptainer, or Pixi for ML stacks.
Diagnoses PyTorch DDP and NCCL failures on AIRE L40S GPU jobs. Use when distributed training hangs, NCCL errors or timeouts occur, only one GPU is active, Slurm resources do not match torchrun, or seff shows wasted GPUs on AIRE.
Scaffolds and maintains modular, reproducible ML research repositories for the AIRE HPC cluster. Use for project layout (src/, configs/, scripts/, tests/), Hydra or YAML configuration, git hygiene, SCRATCH-aware paths, and integrating runs with aire-agent log_experiment. Complements domain skills such as medical-cv-research-engineer.