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/d-robotics/moss/rdk-devicenpx skills add D-Robotics/moss --skill rdk-devicegit clone --depth 1 https://github.com/D-Robotics/mossWrote 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/d-robotics/moss/rdk-device)<a href="https://agentmods.dev/skills/d-robotics/moss/rdk-device"><img src="https://agentmods.dev/badge/skills/d-robotics/moss/rdk-device.svg" alt="Measured on agentmods" height="20"></a>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 | $0.00243 | $0.03392 |
| Opus 5 | $0.00121 | $0.01696 |
| Sonnet 5 | $0.00049 | $0.00678 |
| Haiku 4.5 | $0.00024 | $0.00339 |
Grade C, and why
rdk-device scanned grade C with 2 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 4d 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.
Subtle steeringmediumPrompt injection
Instructions that bias recommendations or shape behaviour without the user noticing.
- ❌ Never tell the user to `apt install hb_mapper` **on the board** — it doesn't belong there. Asks for rootmediumPrivilege escalation
A mod that escalates privileges can change anything on the machine, not only the project.
4. **First config** — `sudo srpi-config` for Wi-Fi / SSH / VNC / locale (not available on Ultra). First boot does ~45s of default setup. How it starts
The opening of the file, as written. The whole thing — 126 lines — stays where its author put it; the contents beside it link to each section on GitHub.
RDK Device Deployment
Take a user's trained model from .pt/.onnx to a working on-board BPU deployment, and get a fresh board booted, networked, and running a camera. The single most common failure is copying a .pt/.onnx straight onto the board — it runs on CPU only and the user blames the hardware. Catch that first, then run the toolchain loop.
Sources: official D-Robotics docs (rdk_doc / rdk_x_doc / rdk_s_doc), the OpenExplorer/天工开物 toolchains, and reproduced community cases. Facts are carried over verbatim with provenance; nothing is invented.
The one rule that matters most
A .pt or raw .onnx does NOT use the BPU. PyTorch / ONNX Runtime on an RDK board run on CPU, so the 40+ TOPS NPU sits idle and the user sees 1–2 FPS. When someone says "I deployed YOLOv5 and it's super slow / video is laggy / FPS won't go up / I copied my pt over" — interrupt before tuning anything and explain they must run the BPU toolchain (.pt → .onnx → .bin/.hbm). Tuning a CPU-path model is wasted effort.
Board → toolchain cheat-sheet (the foundation)
Confirm the board first (cat /sys/class/socinfo/board_id), then everything downstream follows from this table. Cross-architecture artifacts are never interchangeable — a .bin will not load on Nash, and different march values are mutually incompatible.
| Board | BPU arch | march |
Host tool | Artifact | On-board runtime |
|---|---|---|---|---|---|
| RDK X3 | Bernoulli2 | bernoulli2 |
hb_mapper |
.bin |
hobot_dnn (pyeasy_dnn) |
| RDK X5 | Bayes-e | bayes-e |
hb_mapper |
.bin |
hbm_runtime (3.5.0+) / pyeasy_dnn (older) |
| RDK Ultra | Bayes | bayes |
hb_mapper |
.bin |
hobot_dnn |
| RDK S100 | Nash-e | nash-e |
hb_compile |
.hbm |
hbm_runtime |
| RDK S100P | Nash-m | nash-m |
hb_compile |
.hbm |
hbm_runtime |
| RDK S600 | Nash | nash-p |
hb_compile |
.hbm |
hbm_runtime |
march values come from the official FAQ (Super100=Nash-e, Super100P=Nash-m); S600=nash-p confirmed via the LLM SDK's resolve_model_nash-p.md. A quick selector script is in scripts/toolchain_selector.py.
What ships with it
8 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.
- ACCEPTANCE.json 868 B
- assets/templates/s_series_hb_compile_config.yaml 1.1 KB
- assets/templates/x_series_hb_mapper_config.yaml 1.1 KB
- references/board-inference-api.md 16 KB
- references/camera-commands.md 643 B
- references/hardware-notes.md 8.9 KB
- references/toolchain-workflow.md 14 KB
- scripts/toolchain_selector.py 2.9 KB runs code
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.
- 4d ago First seen · 126 lines · 0 tokens per session scan C a1a4d01ab55e
rdk-device is a skill published in the GitHub repository D-Robotics/moss (142 stars, last pushed 8d ago), licensed MIT. It adds 243 tokens to every session and 3,392 once invoked, about $0.0012 per session on Opus 5. A static security scan graded it C with 2 findings (subtle steering, asks for root). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
integrated-browser
Use this when working on the VS Code integrated browser ("browserView") to understand its architecture and mental model. Covers the embedded Chromium browser, its editor tab, navigation, overlay/layout, sessions, and agent browser tools under src/vs/platform/browserView and src/vs/workbench/contrib/browserView.
import-prom-rule
Bulk import of a Prometheus alert rule YAML file (create a whole set of rules at once). Dedicated to handling a remote URL or local YAML text, automatically parsing the three formats groups / a plain rules array / a single rule. ⚠️ Do not use this skill for single-rule creation — when the user describes a single alert…
pcbway
PCBWay PCB fabrication and assembly — turnkey/consigned assembly, design rules, ordering workflow. Alternative to JLCPCB for manufacturing. Use with KiCad. Use this skill when the user mentions PCBWay, needs turnkey assembly (PCBWay sources parts by MPN), has parts not available on LCSC, needs assembled boards with…
unifi-protect
How to manage UniFi Protect cameras and NVR — view cameras, smart detections, Find Anything detection search, recordings, snapshots, lights, sensors, Known Faces, license plates, and the Alarm Manager. Use this skill when the user mentions UniFi cameras, security cameras, NVR, recordings, motion detection, person…
tilelang-env-check
TileLang-Ascend 环境检查与配置验证技能。检查代码仓库完整性、编译安装状态、环境变量配置，并运行简单测试验证环境。发现问题会自动调用相关 skill 进行修复，并按依赖顺序重新执行后续步骤。触发关键词："环境检查"、"检查环境"、"验证环境"、"环境配置"、"环境搭建"、"env check"、"check environment"、"verify environment"、"setup environment"。.
KernelWiki
Use when the user asks about optimizing NVIDIA Blackwell (SM100, B200) or Hopper (SM90, H100) GPU kernels — tcgen05/TMEM/CLC/NVFP4/2-SM cooperative, warp specialization, FlashAttention-4, DeepGEMM, FlashMLA, MoE, grouped GEMM, CuTe-DSL/PTX/Triton on Blackwell, or wants concrete PR references from…