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/axera-tech/magnetar/simulatenpx skills add AXERA-TECH/Magnetar --skill simulategit clone --depth 1 https://github.com/AXERA-TECH/MagnetarWhat 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 | $0.00023 | $0.00759 |
| Opus 5 | $0.00012 | $0.00380 |
| Sonnet 5 | $0.00005 | $0.00152 |
| Haiku 4.5 | $0.00002 | $0.00076 |
Grade A, and why
simulate 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 2d 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.
What it actually says
SIMULATE
执行
metrics = magnetar.stages.simulate.run(task_dir, sample, pulsar_image, board=board, target_hw=TARGET_HARDWARE)
内部逻辑：
- 计算 ONNX 参考输出
- 有板必上板：BOARD 已配置直接用；未配置时先
magnetar.board_util.select_board(TARGET_HARDWARE, BOARD_PASSWORD)找空闲板 - 上板：先
ensure_remote_infer(board)确保 ax-remote-infer 已装（18500 通就跳过，未装自动静默安装），再上传模型，/opt/bin/ax_run_model直接跑（秒级），下载结果与 ONNX 对比 - 仅当找不到板或板端失败：回退
pulsar2 runDocker 仿真（分钟级）
输入/输出格式（bin 命名、input_list、reshape）一律用 magnetar/io_format.py，规范见 docs/input-format-cheatsheet.md。
LLM 分支（model_route=llm）
不再对比 ONNX vs AXMODEL 张量，改为：
- 有板必上板：
magnetar.stages.llm.install_axllm(board)（ax-llm install.sh， 已装则跳过）→serve_axllm(board, compile/llm_model_dir, port=8000)启动 OpenAI 兼容服务 →validate_chat("http://127.0.0.1:8000", model_name, prompts≥3, expected_keyword=None)greedy 语义验证，记录响应非空 / completion_tokens / 耗时；写simulate/simulate_report.md（LLM 版：逐层 cosine + 语义验证指标）； - 无板回退：用
llm_build2 --check_level 2 --prompt <prompt>自带全模型校验，_extract_cosims(log)提取 decode/prefill 逐层 cosine 写入报告；板端语义验证 留到 RUNONBOARD，报告标注 N/A。
验收（accuracy_gate / llm_route_acceptance）：逐层 cosine min ≥ 0.99；
有板时语义验证全非空。
验证
- cosine_similarity ≥ 0.99
- MAE、max_abs_diff 记录在
simulate_report.md - ≥3 组输入样本，报告均值 ± 标准差
- LLM 分支：逐层 cosine min ≥ 0.99 + 有板时 ≥3 组 prompt 语义验证全非空
STOP
- cosine < 0.99：先查
issues/目录已知修复；INT8/U16/混合精度全部尝试仍不达标时，STOP 前先向用户提议上 QAT（必须用官方AXERA-TECH/QAT.axera，走 QAT→QDQ ONNX，需训练数据/时间，用户确认后进入） - AXMODEL 输出全零/异常 → 检查校准归一化配置
- LLM 分支：axllm serve 无法启动 → 检查安装/模型目录/内存；仍失败 STOP
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.
- 2d ago First seen · 45 lines · 23 tokens per session scan A 5e6b9ba6bf49
simulate is a skill published in the GitHub repository AXERA-TECH/Magnetar (22 stars, last pushed 19d ago), licensed MIT. It adds 23 tokens to every session and 759 once invoked, about $0.0001 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
amc-run-rtsp-calibration
Calibrate a new dataset from live RTSP camera streams via the AutoMagicCalib REST API. Use when the user provides RTSP URLs or asks to calibrate live cameras; VIOS records clips, AMC ingests them, then runs calibration.
deepstream-run-mv3dt
Run and operate the DeepStream Multi-View 3D Tracking reference app, also known as MV3DT. Use when the user asks to set up prerequisites, run shipped MV3DT samples, run Multi-View 3D Tracking on custom synchronized MP4 datasets, import camera calibration, delegate missing calibration to AutoMagicCalib, inspect OSD or…
nvidia-skill-finder
Use for NVIDIA-related requests where an NVIDIA skill might help, even if the user did not ask for a skill. Trigger on NVIDIA products, hardware, software, SDKs, GPUs, Jetson/JetPack/L4T/BSP/SDK Manager/driver/flashing/setup, CUDA, NIM, NeMo, Omniverse/OpenUSD/SimReady, RAPIDS/cuDF, cuPyNumeric, cuOpt, Dynamo…
doca-aes-gcm
Use this skill when the user is doing hands-on DOCA AES-GCM work on a BlueField DPU or ConnectX NIC — configuring docaaesgcmtaskencrypt / taskdecrypt, querying docaaesgcmcap for per-key-type (only DOCAAESGCMKEY128 / 256 — AES-192 not supported) and per-task support, sizing plaintext against the max-buf cap, setting…
doca-common
Use this skill whenever the user is doing hands-on DOCA programming on a BlueField DPU or ConnectX NIC and needs the foundation primitives every per-library context rests on — walking the docactx lifecycle, discovering docadev / docadevinfo and gating on docacap before trusting a feature, wiring docammap /…
doca-compress
Use this skill for hands-on DOCA Compress programming on a BlueField DPU, ConnectX NIC, or host with DOCA — enabling compress-deflate, decompress-deflate, decompress-lz4-stream, or decompress-lz4-block tasks on a docacompress context (the hardware supports DEFLATE both directions plus LZ4 decompress; LZ4 encode is NOT…