LLM Torch Profiler Analysis
sgl-project/sglang
Unified LLM torch-profiler triage skill for sglang, vllm, TensorRT-LLM, and TokenSpeed.
Flash the FPGA on an HSB board connected to an NVIDIA devkit.
$ npx skills add NVIDIA/skills --skill hsb-flash -a claude-codeProject install by default; add -g for ~/.claude/skills/.
$ gh skill install NVIDIA/skills hsb-flash --agent claude-codeProject scope by default; add --scope user for a personal install. Needs GitHub CLI 2.90.0 or later (public preview).
$ git clone --depth 1 https://github.com/NVIDIA/skills.git skills-src && mkdir -p .claude/skills && cp -r skills-src/skills/hsb-flash .claude/skills/hsb-flash && rm -rf skills-srcUse ~/.claude/skills/ instead of .claude/skills for a personal install. The folder must contain SKILL.md.
Claude Code skills documentation · loads skills from .claude/skills/
Install the "hsb-flash" agent skill from https://github.com/NVIDIA/skills/tree/main/skills/hsb-flash into .claude/skills/hsb-flash/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "hsb-flash", then confirm the skill loads.Claude Code copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$skill-installer install https://github.com/NVIDIA/skills/tree/main/skills/hsb-flashType this inside Codex. $skill-installer <name> installs a curated skill from openai/skills. The installer writes to $CODEX_HOME/skills (default ~/.codex/skills). Restart Codex if the skill does not show up.
$ npx skills add NVIDIA/skills --skill hsb-flash -a codexProject install goes to .agents/skills/; add -g for ~/.codex/skills/.
$ gh skill install NVIDIA/skills hsb-flash --agent codexProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/NVIDIA/skills.git skills-src && mkdir -p .agents/skills && cp -r skills-src/skills/hsb-flash .agents/skills/hsb-flash && rm -rf skills-srcUse ~/.agents/skills/ instead of .agents/skills for a personal install.
Codex skills documentation · loads skills from .agents/skills/
Install the "hsb-flash" agent skill from https://github.com/NVIDIA/skills/tree/main/skills/hsb-flash into .agents/skills/hsb-flash/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "hsb-flash", then confirm the skill loads.Codex copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add NVIDIA/skills --skill hsb-flash -a cursorProject install goes to .agents/skills/; add -g for ~/.cursor/skills/.
$ gh skill install NVIDIA/skills hsb-flash --agent cursorProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/NVIDIA/skills.git skills-src && mkdir -p .cursor/skills && cp -r skills-src/skills/hsb-flash .cursor/skills/hsb-flash && rm -rf skills-srcUse ~/.cursor/skills/ instead of .cursor/skills for a personal install.
Cursor skills documentation · loads skills from .cursor/skills/, .agents/skills/, .claude/skills/, .codex/skills/
Install the "hsb-flash" agent skill from https://github.com/NVIDIA/skills/tree/main/skills/hsb-flash into .cursor/skills/hsb-flash/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "hsb-flash", then confirm the skill loads.Cursor copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gemini skills install https://github.com/NVIDIA/skills.git --path skills/hsb-flash--scope user (default) or --scope workspace; --path is the subfolder of the repo that holds the skill; --consent skips the security confirmation prompt.
$ npx skills add NVIDIA/skills --skill hsb-flash -a gemini-cliProject install goes to .agents/skills/; add -g for ~/.gemini/skills/.
$ gh skill install NVIDIA/skills hsb-flash --agent gemini-cliProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/NVIDIA/skills.git skills-src && mkdir -p .gemini/skills && cp -r skills-src/skills/hsb-flash .gemini/skills/hsb-flash && rm -rf skills-srcUse ~/.gemini/skills/ instead of .gemini/skills for a personal install, then run /skills reload.
Gemini CLI skills documentation · loads skills from .gemini/skills/, .agents/skills/
Install the "hsb-flash" agent skill from https://github.com/NVIDIA/skills/tree/main/skills/hsb-flash into .gemini/skills/hsb-flash/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "hsb-flash", then confirm the skill loads.Gemini CLI copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ gh skill install NVIDIA/skills hsb-flashInstalls for Copilot at project scope by default; add --scope user for a personal install. Preview a skill first with gh skill preview. Needs GitHub CLI 2.90.0 or later (public preview).
$ npx skills add NVIDIA/skills --skill hsb-flash -a github-copilotProject install goes to .agents/skills/; add -g for ~/.copilot/skills/.
$ git clone --depth 1 https://github.com/NVIDIA/skills.git skills-src && mkdir -p .github/skills && cp -r skills-src/skills/hsb-flash .github/skills/hsb-flash && rm -rf skills-srcUse ~/.copilot/skills/ instead of .github/skills for a personal install. Commit .github/skills so cloud agent and code review can use it.
GitHub Copilot skills documentation · loads skills from .github/skills/, .claude/skills/, .agents/skills/
Install the "hsb-flash" agent skill from https://github.com/NVIDIA/skills/tree/main/skills/hsb-flash into .github/skills/hsb-flash/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "hsb-flash", then confirm the skill loads.GitHub Copilot copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
$ npx skills add NVIDIA/skills --skill hsb-flash -a opencodeOpenCode documents no install command of its own. Project install goes to .agents/skills/; add -g for ~/.config/opencode/skills/.
$ gh skill install NVIDIA/skills hsb-flash --agent opencodeProject scope by default (.agents/skills/); add --scope user for a personal install.
$ git clone --depth 1 https://github.com/NVIDIA/skills.git skills-src && mkdir -p .opencode/skills && cp -r skills-src/skills/hsb-flash .opencode/skills/hsb-flash && rm -rf skills-srcUse ~/.config/opencode/skills/ instead of .opencode/skills for a personal install.
OpenCode skills documentation · loads skills from .opencode/skills/, .claude/skills/, .agents/skills/
Install the "hsb-flash" agent skill from https://github.com/NVIDIA/skills/tree/main/skills/hsb-flash into .opencode/skills/hsb-flash/ in this project. Copy the whole folder (SKILL.md and every file beside it), keep the folder name "hsb-flash", then confirm the skill loads.OpenCode copies the folder itself, the same result as the manual copy. Check what it changed before you commit it.
hsb-flashFlash the FPGA on an HSB board connected to an NVIDIA devkit.
Hsb Flash is an agent skill from NVIDIA/skills, published by the product's own GitHub organization. Flash the FPGA on an HSB board connected to an NVIDIA devkit. Supports HSB Lattice boards (FPGA versions 2407, 2412, 2507, 2510) and Leopard Imaging VB1940 "all-in-one" cameras (FPGA versions 2507, 2510). Uses release-specific YAML manifests and board-type-specific program commands. Lattice and VB1940 commands must never be mixed.
Its SKILL.md is about 4.4k tokens, which your agent loads only when the skill is triggered. The skill folder holds 23 other files, including scripts and reference files (for example `BENCHMARK.md`, `evals/evals.json` and `references/flashing-infrastructure.md`).
It works with NVIDIA AI Platform. The repository describes itself as: Agent Skills for NVIDIA products — install into Claude Code, Codex, and other coding agents to run Physical AI, robotics, simulation, CUDA, and RAG workflows end to end. The licence is Apache-2.0.
2 steps, taken from the first numbered list in SKILL.md.
Read from SKILL.md and the folder at commit 0e0d506. It shows what the files ask for, not the result of running them.
Pre-approves these tools, so the agent can use them without asking each time:
ReadWriteEditMultiEditGrepGlobBashFrom allowed-tools in the SKILL.md frontmatter.
Ships 8 files in scripts/ (Python, from the files we listed), which the agent can run.
From the folder's file list and the shell code blocks in SKILL.md.
Hosts in commands or code, which the agent is likely to contact:
github.comFrom URLs in SKILL.md, links to its own repository left out.
Names no API keys, tokens, secrets or passwords.
From names ending in _API_KEY, _TOKEN, _SECRET, _KEY or _PASSWORD in SKILL.md.
Hsb Flash loads about 4.4k tokens when it runs, and up to ~27k if it reads all its reference files. Until then it costs about 86 tokens; SKILL.md has 2,315 words of instructions outside code blocks.
Estimates: characters ÷ 4, the usual rule of thumb; real counts depend on the model's tokenizer. Scripts and assets cost tokens only if the agent reads them.
The automated check noted patterns worth knowing about, such as sudo or a known installer.
REMOTE_SUDO sudo / sudo -n / "" — default to "sudo" if not set.ecuted **inside the demo container** (no sudo needed within the container).allowed-tools: Read, Write, Edit, MultiEdit, Grep, Glob, BashAutomated static check — not a guarantee. Review scripts before installing. It scans the text of SKILL.md for risky patterns (piping downloads into a shell, reading credential files, hidden Unicode, destructive commands); the scripts in this folder are not scanned.
The full file from NVIDIA/skills at commit 0e0d506, republished under its Apache-2.0 licence (© NVIDIA). 2,315 words, ~4,413 tokens.
.claude/skills/hsb-flash/SKILL.md (or your agent's skills folder). This skill also uses 16 other files; get the full folder from GitHub.Use this skill when the user wants to flash (upgrade or downgrade) the FPGA firmware on an HSB board connected to a supported NVIDIA devkit.
This skill supports two board types:
CRITICAL SAFETY RULE: Never mix board-type commands. Using program_leopard_cpnx100 on a Lattice board or program_lattice_cpnx100 on a VB1940 can permanently brick the device. The skill must detect and confirm the board type before any flash operation, and refuse to proceed if the board type is ambiguous or mismatched.
This workflow has side effects (it permanently modifies FPGA firmware). Never run it automatically. Only run it when the user explicitly invokes it.
Usage warning: This skill flashes the FPGA with new firmware. Before invoking it, ask the user to make sure they have enough Claude Code usage/tokens to complete the workflow.
Gate 1 — Read environment variables. Before doing anything else, check these variables and print their resolved values to the user:
SSH_TARGET Remote devkit login (e.g. nvidia@192.168.1.50). Ask the user if not set.
REMOTE_ROOT Remote working directory (e.g. /home/nvidia). Ask the user if not set.
REMOTE_SUDO sudo / sudo -n / "" — default to "sudo" if not set.
REMOTE_SSH_OPTS Additional SSH options (optional).
HSB_PLATFORM Platform hint (optional).SSH_TARGET and REMOTE_ROOT are required. Stop and ask the user for them if either is missing.
Gate 2 — Present the flash summary and phase plan. Before taking any action:
If the user's request already includes board type, current FPGA version, and target FPGA version, state the following before the phase plan: flash tool (program_lattice_cpnx100 for Lattice, program_leopard_cpnx100 for VB1940 — never mix), manifest release and filename, CLI flags (--force --accept-eula), whether the procedure is single-step or two-step via gateway 2412. For VB1940, also state that no v2.0.0 interim repo is needed. For two-step upgrades from FPGA 2407, state that step 1 uses hololink --force fpga_version (not hololink enumerate, which is incompatible with FPGA 2407) and uses v2.0.0 flag placement: hololink --force program scripts/manifest.yaml --accept-eula (--force before the subcommand).
Then show the phase plan and ask explicitly: Shall I proceed with the flash workflow? [Y/n] — do not start Gate 3 until the user confirms:
HSB Flash — Phase Plan
Phase 0: Token-budget preflight
Phase 1: Verify board connectivity, detect board type (Lattice or VB1940), read FPGA version
Phase 2: Select target FPGA version
Phase 3: Prepare flash infrastructure and YAML files, present flash plan for approval
Phase 4: Execute flashing procedure (with power cycle verification)
Phase 5: Summary report (with option to save)
Phase 6: Clean up flash artifactsGate 3 — Token-budget preflight (Phase 0). Run after the phase plan (Gate 2) has been presented and the user has confirmed. Do not run the token-budget check before the phase plan is shown. Do not proceed to Phase 1 until the budget check passes.
Gate 4 — Confirm board type explicitly. Before any flash command, confirm with the user whether the board is Lattice or VB1940. Never mix program_lattice_cpnx100 and program_leopard_cpnx100 — wrong tool can brick the device.
Invoke this skill by typing /hsb-flash [OPTIONS]. The skill detects the board type automatically, presents a flashing plan, and prompts for confirmation before each flash step. See references/help-text.md for the full --help output.
hololink enumerate first; if it fails (which is expected for FPGA 2407 boards), fall back to hololink --force fpga_version. For Lattice boards, if all methods fail with the existing repo's container, checkout HSB release repo v2.0.0 and retry using the v2.0.0 container. If that also fails, assume the version is 2407 and continue. For VB1940 boards, ask the user if the version cannot be read.https://github.com/nvidia-holoscan/holoscan-sensor-bridge/blob/main/RELEASE_NOTES.md and look for a release that introduces the undocumented FPGA version. If a matching release is found, checkout that release repo on the devkit and use it for flashing following the same rules described below for the detected board type. Also update this skill's mapping tables, supported FPGA versions lists, and transition matrices with the new release and its corresponding FPGA version./hsb-setup) to flash, following the same rules for the detected board type. If the flash fails, report the error and prompt the user for further instructions.scripts/ directory of this skill to the checked-out repo, and patch the file as needed for any missing details (e.g., fpga_uuid).--force and --accept-eula to ensure non-interactive execution inside the container. Note: v2.0.0 places --force before the subcommand and --accept-eula after — see "v2.0.0 CLI flag placement" below.| Board Type | Identifier | Description |
|---|---|---|
| Lattice | lattice | HSB Lattice CPNX100-ETH-SENSOR-BRIDGE standalone FPGA board |
| VB1940 | vb1940 | Leopard Imaging VB1940 "all-in-one" Eagle Camera with integrated Lattice FPGA |
The board type is detected from the hololink enumerate output during Phase 1 and confirmed with the user. If detection is ambiguous, the user must explicitly specify the board type.
| Version | YAML Source Release | Notes |
|---|---|---|
| 2407 | v2.0.0 | Oldest supported version |
| 2412 | v2.0.0 | Gateway version for two-step flashing |
| 2507 | v2.3.1 | |
| 2510 | v2.5.0 | Latest supported version |
| Version | YAML Source Release | HSB Release | Notes |
|---|---|---|---|
| 2507 | v2.3.0 | v2.3.0 | |
| 2510 | v2.5.0 | v2.5.0 | Latest supported version |
The VB1940 does not support versions 2407 or 2412 — these are Lattice-only.
Versions not listed above: FPGA versions newer than the latest documented version for either board type may still be flashable — see "Handling undocumented FPGA versions" below. For Lattice boards, versions older than 2407 or between known versions (e.g., 2409) are not supported. For VB1940, versions older than 2507 are not supported. In either case, refuse and show the supported versions for the board type.
See references/flashing-infrastructure.md for GitHub release tags, bundled manifest YAML layout, board-specific flash commands, v2.0.0 CLI flag differences, and FPGA 2407 enumerate workaround.
The "Lattice board FPGA versions" table above determines which HSB release repo to use. The lookup key depends on direction:
Self-updating: If an undocumented FPGA version is encountered, the skill checks the public release notes for a matching HSB release (see "Handling undocumented FPGA versions"). If found, the skill updates the "Supported FPGA versions" tables, the transition matrix, and notes the new release's manifest files.
/hsb-setup), read its version from the VERSION file.VB1940 note: VB1940 flashing always uses the existing repo on the devkit — the FPGA-to-repo mapping does not apply. The existing repo must be at version v2.3.0 or later. If no existing repo is found, instruct the user to run
/hsb-setupfirst.
During Phase 1, when scanning for an existing repo and detecting the board type, save these variables to the session state:
BOARD_TYPE — the detected board type: lattice or vb1940EXISTING_REPO_DIR — absolute path to the existing HSB repo (empty if none found)EXISTING_REPO_VERSION — the repo's release version (e.g., 2.3.1), read from the VERSION fileFLASH_REPO_DIR — absolute path to the repo that will be used for flashing (may differ from EXISTING_REPO_DIR if a different version was checked out)FLASH_REPO_VERSION — the version of the flash repo (looked up from the FPGA-to-repo mapping)INTERIM_REPOS — list of repo directories checked out by this skill (for cleanup in Phase 6)Reuse the same environment variables from the hsb-setup skill:
SSH_TARGET for the remote login target (e.g. nvidia@agx-thor-host)REMOTE_ROOT for the remote working directory where flash workspace will be createdREMOTE_SUDO for privileged commandsREMOTE_SSH_OPTS for additional SSH optionsHSB_PLATFORM as an optional platform hintIf these are set, notify the user of these settings and use them without re-asking.
Before Phase 1, print the resolved remote execution settings.
Before making changes, show this phase plan:
Then execute one phase at a time.
After each non-final phase (Phases 0–5):
Proceed to Phase <N+1>? [Y/n] and specify what the next phase does. Wait for confirmation before continuing.Exception: When --y (auto-approve mode) is active, phase gates are skipped and phases run automatically. See "Auto-approve mode (--y)" section for details.
If something fails, do not just dump raw logs. Summarize:
See references/phase-details.md for full step-by-step phase instructions, flashing procedure logic, execution rules, safety constraints, phase gate rules, verbosity behavior, force mode, and auto-approve mode.
--help)See references/help-text.md for the full --help output text.
See references/help-text.md for the full --help output including all invocation examples.
© NVIDIA, Apache-2.0. Rendered from Markdown: HTML in the file is shown as text, images as links, and headings moved down two levels. Raw file
SKILL.md and 16 other files (scripts, references) in skills/hsb-flash of NVIDIA/skills.
Open the folder on GitHubat commit 0e0d506
Hsb Flash next to the 5 skills that share the most tags, products or categories with it. Stars are the repository's; “used in” counts other GitHub owners with a copy.
| Skill | Stars | Used in | Tokens | Auto-check | Licence | Repo updated |
|---|---|---|---|---|---|---|
| Hsb Flash this skillNVIDIA/skills | 3.5k | — | ~4.4k | Automated safety check: Notes | Apache-2.0 | |
| LLM Torch Profiler Analysissgl-project/sglang | 37k | 2 repos | ~6.4k | Automated safety check: Pass | Apache-2.0 | |
| Skill InspectorNVIDIA/SkillSpector | 20k | — | ~1.8k | Automated safety check: Pass | Apache-2.0 | |
| Embeddings via 9Routerdecolua/9router | 30k | — | ~604 | Automated safety check: Pass | MIT | |
| NEAR AI Cloud Private Inferenceinternet-court/internet-court-skill | 6.4k | 2 repos | ~1.3k | Automated safety check: Pass | Custom licence | |
| Nemoclaw Maintainer Normalize Title TagsNVIDIA/NemoClaw | 23k | — | ~693 | Automated safety check: Pass | Apache-2.0 |
sgl-project/sglang
Unified LLM torch-profiler triage skill for sglang, vllm, TensorRT-LLM, and TokenSpeed.
NVIDIA/SkillSpector
Decides whether an agent skill is safe to install by combining a SkillSpector static scan with the agent's own source review, ending in APPROVE, CAUTION or REJECT.
decolua/9router
Generates vector embeddings through the 9Router /v1/embeddings endpoint, using models from providers such as OpenAI, Gemini, Mistral and Voyage for RAG and semantic search.
internet-court/internet-court-skill
Shows how to call NEAR AI Cloud through an OpenAI-compatible API and verify that inference ran in a TEE, using attestation checks and signed chat responses.
NVIDIA/NemoClaw
Remove bracketed NemoClaw tags from GitHub issue and PR titles.
NVIDIA/Megatron-LM
Walks an agent through working inside the Megatron-LM CI container and changing dependencies with uv, so lock files resolve the same locally and in CI.
NVIDIA/skills
A skill your agent uses when the user wants to deploy, run, debug, tear down, or call the REST API of the RTVI-CV 2D detection / tracking microservice.
NVIDIA/skills
Generates, validates, compares and explains HOLOLINK_def.svh macro files for the HSB IP, using bundled Python scripts and asking before it writes anything.
NVIDIA/skills
Runs and validates an end-to-end Mission Control demo in a locally installed Isaac Sim, with a Nova Carter robot driven through a Python server.
NVIDIA/skills
Orchestrates defect image generation for PCBA, metal surface and glass inspection with NVIDIA Cosmos AnomalyGen on OSMO, from cold-start Day 0 to real-photo Day 1 labeling.
NVIDIA/skills
Orchestrates video data augmentation and auto-labeling workflows on OSMO, from flow selection and preflight checks to submission, monitoring and output download.
NVIDIA/skills
Runs NVIDIA TAO Data Services KPI analysis on object detection results, comparing predictions to ground truth and writing per-class precision, recall and AP to a CSV.
Works with
Flash the FPGA on an HSB board connected to an NVIDIA devkit. Hsb Flash is an agent skill from NVIDIA/skills, published by the product's own GitHub organization. Flash the FPGA on an HSB board connected to an NVIDIA devkit.
Run `npx skills add NVIDIA/skills --skill hsb-flash -a claude-code`. Or copy the skill folder (skills/hsb-flash in NVIDIA/skills) into .claude/skills/hsb-flash in your project. Claude Code loads it when a task matches its description.
Run `npx skills add NVIDIA/skills --skill hsb-flash -a codex`. Or copy the skill folder (skills/hsb-flash in NVIDIA/skills) into .agents/skills/hsb-flash in your project. Codex loads it when a task matches its description.
Cursor, Gemini CLI, GitHub Copilot and OpenCode also load SKILL.md folders. With the skills CLI, run `npx skills add NVIDIA/skills --skill hsb-flash -a cursor` (or -a gemini-cli, github-copilot or opencode for the others). To copy it by hand, put the folder in .cursor/skills/hsb-flash, .gemini/skills/hsb-flash, .github/skills/hsb-flash and .opencode/skills/hsb-flash in your project.
Going by SKILL.md and its folder, Hsb Flash needs Python for the scripts in its folder. Our summary lists: Python 3. Its frontmatter pre-approves these tools: Read, Write, Edit, MultiEdit, Grep, Glob, Bash.
SKILL.md names 1 domain. In commands or code: github.com; the agent is likely to contact it when it follows the instructions. This is read from the text; nothing was executed.
Our automated static check of SKILL.md found notes only (runs commands with sudo; pre-approves every shell command (allowed-tools: bash)), nothing it rates as a warning. It is not a guarantee. The check reads SKILL.md only: the scripts in the folder are not scanned, so read them before running anything.
Hsb Flash is published under the Apache-2.0 licence (declared in SKILL.md). It allows redistribution, so the full SKILL.md is shown on this page.
About 4.4k tokens (SKILL.md is roughly 18k characters). Agents keep only the skill's name and description in context until a task matches; then they load SKILL.md in full. Its references folder adds about 22k tokens, read only when the agent opens those files.
Skills that share tags, products or a category with Hsb Flash: LLM Torch Profiler Analysis (sgl-project/sglang, 37k stars), Skill Inspector (NVIDIA/SkillSpector, 20k stars), Embeddings via 9Router (decolua/9router, 30k stars) and NEAR AI Cloud Private Inference (internet-court/internet-court-skill, 6.4k stars). The comparison table on this page puts their stars, adoption, token cost, safety result and licence side by side.
NVIDIA (a GitHub organization, an official publisher) maintains it in NVIDIA/skills, which has 3,534 GitHub stars. The repository holds 380 skills in this directory. The repository was last updated on October 7, 2026.
Source: NVIDIA/skills on GitHub. Facts on this page come from the repository at the commit we read; the author's words are quoted as theirs.