Jetson BSP Image Initializer
Prepares a Jetson BSP image, applies binaries, and records its verified version.
The skill scopes changes to the active profile's bsp_image block, requests confirmation for defaults or overrides, refuses unknown modules, version mismatches, and unconfirmed overwrites, and explicitly does not download tarballs. However, it runs sudo, extracts user-supplied archives, and modifies the filesystem without archive integrity/source-authenticity checks, backup or rollback, or sufficiently explicit root-privilege data-flow disclosure, so points are deducted.
The workflow documents GPU-stack selection, reuse detection, version conflicts, and failure guidance, and includes two evaluated tasks with reported results. However, there is no executable implementation or committed test suite for this skill; it depends on tar, sudo, ruamel.yaml, and BSP file formats, while the command excerpt does not show explicit set -e or transactional recovery. Static evidence is therefore insufficient and the score is conservatively limited.
Invocation triggers, prerequisites, tarball inputs, profile target, and the non-fit source-tree setup are clear, with separate paths for Orin and Thor+. Environment compatibility, Chinese interaction, and mainland-China reachability are not addressed, and only a small set of module IDs is covered, so points are deducted.
SKILL.md provides metadata, layered procedures, limitations, troubleshooting, references, and a version; repository context provides licensing, official organization provenance, and daily mirror maintenance signals. It lacks the recommended Instructions and Examples sections, the author format fails the benchmark requirement, the skill card's CC-BY-4.0 AND Apache 2.0 statement conflicts with the Apache-2.0 frontmatter, and changelog and explicit maintenance ownership are incomplete, so points are deducted.
The goal is concrete and the documented workflow connects matching BSP/rootfs extraction, apply_binaries.sh, and profile persistence into a directly usable procedure. The benchmark reports 94% Codex correctness and 59% effectiveness, but it contains only two positive tasks and no negative tasks; static review cannot verify actual artifacts, round-tripping YAML behavior, or hardware-specific results, so the score is conservatively reduced.
The fixed revision, evaluation tasks, ground truth, behavior assertions, and benchmark metrics provide limited auditable evidence. However, there is no committed test suite or real CI coverage for this skill and no independently reproducible output; the benchmark only reports an external run, leaving key claims primarily document-based, so points are deducted.
- Execution includes sudo, sample-rootfs extraction, and apply_binaries.sh; manually verify archive provenance, integrity, target paths, and GPU stack before running.
- No automatic backup or rollback is specified; preserve the profile and Image state before execution even though overwrite safeguards exist.
- License wording should be reconciled: the frontmatter says Apache-2.0 while the skill card says CC-BY-4.0 AND Apache 2.0.
- The evaluation contains only two positive tasks and no negative-trigger tests, so false-positive avoidance is not well established.
What does this skill do, and when should you use it?
This skill prepares a BSP image for an already resolved Jetson target. It extracts matching Jetson Linux and sample-rootfs archives, runs apply_binaries.sh, and records the derived BSP version in the active target profile. It selects nvgpu or OpenRM from the module ID and checks existing images for stack consistency. It owns only the bsp_image block; it does not download archives or set up a source tree.
Reads the active profile's reference_devkit.module.id and bsp_image state; resolves the default or overridden image path; locates matching Jetson Linux and sample-rootfs .tbz2 archives; extracts Linux_for_Tegra and the root filesystem; runs apply_binaries.sh with --openrm when required and supported; derives the version from nv_tegra_release; probes an existing image's GPU stack; updates the active profile's bsp_image block; and reports the image path, reused or extracted state, GPU stack, version, and profile-update status.
- A Jetson BSP user has staged both matching archives and wants to initialize the default Image directory.
- An Orin/T234 user needs the nvgpu stack applied to an extracted BSP.
- A T264-or-later user needs OpenRM applied when the active BSP exposes the --openrm flag.
- A user with an existing Linux_for_Tegra directory wants its version and GPU stack checked and recorded in the target profile.
What are this skill's strengths and limitations?
- Checks that both archive filenames carry the same R<ver> token.
- Derives nvgpu or OpenRM from the module ID and verifies an existing image where possible.
- Persists the derived BSP version in the active profile for downstream workflows.
- Refuses unknown modules, stack conflicts, and unconfirmed destructive replacement.
- Does not download BSP or rootfs archives; inputs must be staged beforehand.
- Requires shell execution, tar, sudo, filesystem write access, and profile-writing support.
- Does not handle source-tree setup, carrier customization, or flashing.
- The supplied material provides no standalone test suite, platform coverage matrix, or performance data.
How do you install this skill?
Install the individual skill through the NVIDIA collection CLI: npx skills add nvidia/skills --skill jetson-init-image --yes. The README does not specify the final installation directory; the CLI handles the destination.
How do you use this skill?
Resolve an active target and stage matching Jetson Linux and sample-rootfs archives, then ask the agent to “extract and initialize the Jetson BSP image.” An active profile with reference_devkit is required. Archives are searched under the image root, workspace, and current working directory; missing files require absolute paths. The workflow needs write access to the image root or configured override and requires sudo. After completion, it suggests /jetson-init-source.