AutoGineer
Generated banner for Jetson Customize Camera
Generated logo for Jetson Customize Camera

Jetson Customize Camera

NVIDIAOfficialCreated by Jetson TeamDistributed by NVIDIA

  • Skill
  • Workflow
  • Software development
  • Apache-2.0
  • Medium risk
Download ZIP

From the author

Enable MIPI/GMSL camera sensors on a Jetson Thor or Orin custom carrier by rendering a kernel-DT overlay from the in-tree sensor DTSI. Do NOT use for UPHY lane allocation or ODMDATA edits.
— the description by Jetson Team

AutoGineer summary

Written by AutoGineer from the skill’s own files, not by the author

This skill automates the bring-up of MIPI and GMSL camera sensors on custom Jetson Thor or Orin carrier boards. It functions by rendering a kernel-DT overlay from in-tree sensor DTSI files, ensuring correct wiring and configuration without requiring manual ODMDATA or UPHY lane adjustments. The process is agentic, utilizing a structured workflow to verify carrier and module support, generate the necessary device tree fragments, and validate the final composite overlay before compilation.

What it does

  • Enumerates supported sensors from in-tree DTBO files.
  • Cross-references carrier and module support against official documentation.
  • Generates and appends a single camera-tagged fragment to the composite custom overlay.
  • Verifies ancillary pin configurations like I2C, clocks, and GPIOs.
  • Validates the final device tree structure using dtc and fdtoverlay.
  • Automates the commit process for the hardware repository.

When to use it

  • When configuring a new MIPI or GMSL camera sensor on a custom Jetson carrier board.
  • When a camera is physically connected but fails to enumerate in v4l2-ctl.
  • When adding additional sensors to an existing multi-sensor Jetson platform.
  • When you need to ensure correct CSI port plumbing and sensor driver binding.

Advantages

  • Reduces manual errors by using in-tree DTSI files as the source of truth.
  • Prevents common pitfalls like dual-fragment conflicts or incorrect port indexing.
  • Maintains a clean, idempotent workflow for hardware customization.
  • Ensures compatibility with the existing BSP and flash-conf registration.

Usage examples

“Enable the IMX274 camera on my custom Orin carrier board.”
The agent identifies the sensor, verifies the carrier pinmap, generates the required kernel-DT overlay fragment, and commits it to the BSP source repository.
“I need to add a GMSL camera to my existing Jetson Thor setup.”
The agent performs a multi-sensor bring-up check, splices the necessary GMSL serdes configuration into the composite overlay, and validates the new endpoint connections.

Install

  1. Download the skill as a ZIP:

    https://mp-staging.autogineer.com/agent-skills/nvidia/jetson-customize-camera/download

  2. Unzip it into the cross-client skills folder of your project:

    mkdir -p .agents/skills && unzip jetson-customize-camera.zip -d .agents/skills

Documentation by Jetson Team

Customize camera (CSI / MIPI / GMSL sensor bring-up)

Overview

Tegra264 (Thor) and Tegra234 (Orin) expose a single tegra-capture-vi controller fronted by NVCSI and a fixed set of CSI ports. Camera bring-up is:

  1. Sensor selection — picked from the set NVIDIA ships in-tree .dtsi references for on the active platform.

  2. Carrier + module support check — verified against the Camera Development Guide, Adaptation Guide §Camera, carrier schematic, Module TRM, and carrier pinmap.

  3. Wiring — derived from the in-tree tegra<soc>-camera-<sensor>*.dtsi when one exists (the DTSI IS the wiring source of truth); captured per-sensor from the user when the sensor is custom.

  4. Kernel-DT overlay — cpp-expand the in-tree DTSI, extract its fragment@N body, append into the composite custom overlay .dts for the active target (per ../../references/bsp-customization-kernel-dtb.md), verify the composite with fdtoverlay. /jetson-build-source compiles the composite and owns the carrier conf's OVERLAY_DTB_FILE+= registration.

Agentic, not table-driven — sensor list is built at runtime by globbing in-tree per-sensor dtbos. No _THOR_CAMERAS dict, no questions.json, no Python renderer in the question path.

No ODMDATA edit — cameras don't consume UPHY lanes (CSI is a separate PHY pool). The skill emits only a kernel-DT overlay; the ODMDATA line in the carrier conf is untouched by this skill.

The output is one commit:

  • Camera fragment@N block (plus jetson-header-name on the composite root if not already present) appended to the composite custom overlay .dts per ../../references/bsp-customization-kernel-dtb.md → committed to the bsp_sources/ hardware repo. /jetson-build-source compiles the composite to .dtbo and owns its Makefile + flash-conf registration.

When to invoke

  • The user says "enable camera", "configure CSI", "wire a Hawk / Owl / IMX sensor", "MIPI camera", "GMSL camera", or asks to bring up tegra-capture-vi / NVCSI on a custom carrier.

  • Flash boots but v4l2-ctl --list-devices shows no tegra-capture-vi channels, OR sensor enumeration on a fresh daughter-card needs to be confirmed.

  • A sensor was previously enabled and the user wants to add another (multi-sensor bring-up).

Prerequisites:

  • Active profile with reference_devkit: + custom_carrier: blocks.

  • <source.root_path>/Linux_for_Tegra/.git exists (/jetson-init-source).

  • /jetson-derive-carrier has run — the carrier flash-conf fork is in the overlay tracker.

  • <source.root_path>/bsp_sources/hardware/nvidia/<chip-dir>/nv-public/overlay/ exists and contains the in-tree per-sensor .dtsi files (sourced by /jetson-init-source's Branch A archive extract).

  • <source.root_path>/bsp_sources/kernel/kernel-noble/include/dt-bindings/ contains the macro headers cpp needs (source_sync.sh may need to run if Branch B was used — see Step 5a.i below).

  • Source-of-truth docs registered or supplied at prompt: Camera Development Guide (in bsp_developer_guide mirror or separate path), Adaptation Guide §Camera, carrier schematic, SoC TRM, Module Design Guide.

  • dtc, cpp, fdtoverlay on PATH.

Procedure

Detailed step-by-step procedure (Steps 1–7, with all tables, code blocks, and gates) lives in references/procedure.md. Summary:

  1. Step 1 — Resolve active target + open source-of-truth docs.

  2. Step 2 — Enumerate supported sensors by globbing in-tree per-platform camera dtbos; classify as DPHY-direct / GMSL / custom. Never invent sensors.

  3. Step 3 / 3a — Cross-check carrier + module support against DTSI, Camera Development Guide, Adaptation Guide §Camera, SoC TRM, Module Design Guide, schematic, and carrier pinmap. Render the wiring table FIRST, then issue the confirm-or-customize gate.

  4. Step 4 (custom path only) — Batched per-sensor wiring questions auto-filled from the carrier pinmap.

  5. Step 5 — Append exactly ONE /* custom-bsp: camera:<sensor> */ fragment to the composite custom overlay .dts (see ../../references/bsp-customization-kernel-dtb.md). Clone path cpp-expands the in-tree DTSI; custom path splices Step-4 answers + mode tables in-place. Idempotently set jetson-header-name on the composite root. Verify with dtc + fdtoverlay (pre-compile single-fragment gate; post-compile deep-tree uniqueness gate). Commit via the workflow's commit-message preview gate.

  6. Step 6 — Verify ancillary CAM pin SFIOs (cam_i2c_*, extperiph<m>_clk, reset/PWDN/PWR_EN GPIOs) via pin_verifier.py; route mismatches to /jetson-customize-pinmux.

  7. Step 7 — Atomic-write run-state JSON sidecar at <workspace>/target-platform/<profile-stem>.jetson-customize-camera.json and emit the headline, then drive the downstream next-step chain via sequential AskUserQuestion prompts per references/procedure.md Step 7. The chain is a documented workflow gate, not a clarifying question — auto-mode does NOT exempt it. Never substitute a printed "Next step: …" line for the prompts.

Gotchas

  • Dual-fragment trap. Contribute exactly ONE camera-tagged fragment@N to the composite. A second one carrying status overrides triggers dtc deep-merge → duplicate sibling subtrees (e.g. two tca9546@70) → runtime first-match drops the dtsi- supplied deep tree → camera silently doesn't enumerate. Gate on this skill's marker only (Step 5c).

  • Composite root compatible is owned globally, not by this skill. Don't widen from any in-tree per-sensor dtbo's compatible (devkit-SKU-gated). Fix the composite root if needed.

  • jetson-header-name from any in-tree per-sensor dtbo. Fixed, carrier-agnostic; read once, paste onto the metadata root.

  • DO NOT also append the in-tree per-sensor dtbo to OVERLAY_DTB_FILE. Registering both your rendered overlay AND the in-tree tegra<soc>-p3971-camera-<sensor>-overlay.dtbo produces a phantom subdev bind that bricks camera enumeration.

  • Stub overlay is a known footgun. Committing tegra-capture-vi { status="okay"; num-channels=<N>; } with no ports / sensor / nvcsi body bricks the camera (all channel init failed). Splice the FULL sensor body via cpp + dtc.

  • Sensor mode tables must be spliced, never hand-authored. mode<N>, sensor_modes, pixel_phase — copy verbatim from the closest in-tree DTSI.

  • camera_common_regulator_get (null) ERR: -EINVAL = missing avdd-reg / iovdd-reg / dvdd-reg strings — splice the FULL sensor body; always-on rails fall back to dummy regulator.

  • External &label refs must exist in base DTB's __symbols__. Use target-path = "/tegra-capture-vi" when the label is absent; fdtoverlay exits non-zero with FDT_ERR_NOTFOUND otherwise.

  • cpp failure on dt-bindings/gpio/gpio.h: No such file = L4T source tree isn't staged. Re-run /jetson-init-source (Branch B's source_sync.sh fetches the headers). Never fabricate the macro expansion.

  • No ODMDATA edit, no flash-conf edit. Camera doesn't consume UPHY lanes. The carrier conf's ODMDATA="..." is untouched. OVERLAY_DTB_FILE+= is owned by /jetson-build-source Step 5.0a — this skill never touches the carrier flash conf.

  • Don't touch the upstream BSP at <bsp_image.root_path>. All edits land in <source.root_path>/Linux_for_Tegra/ (overlay tracker) and <source.root_path>/bsp_sources/ (overlay .dts) under the pristine + customization commit pattern.

References

  • references/procedure.md — full step-by-step Steps 1–7 procedure (extracted from this SKILL.md).

  • references/csi-dt-bindings.md — CSI / nvcsi / vi DT binding reference notes.

  • references/overlay-template.md — guidance on the metadata-root + clone-body overlay shape.

  • references/camera-overlay-templates/ — starter .dts.tmpl templates: dphy-direct.dts.tmpl, gmsl-serdes.dts.tmpl.

  • ../../scripts/pin_verifier.py — shared HSIO pin verifier (Step 6).

  • ../../references/platform_template.yaml — documents: block consumed by Step 1.

  • ../../context/bsp-customization-workflow.md — overlay edit protocol.

  • ../jetson-customize-pinmux/SKILL.md — sibling skill auto-invoked by Step 6 to fix HSIO pin SFIO mismatches (CAM I²C, MCLK, reset GPIOs).

  • ../jetson-derive-carrier/SKILL.md — must run first; produces the carrier base overlay (the *-dynamic.dtbo) this skill's composite stacks after.

  • ../jetson-init-source/SKILL.md — produces the overlay tracker + bsp_sources repo (with the hardware/nvidia/<chip-dir>/ per-sensor DTSI tree) this skill reads and commits into.

Files

BENCHMARK.md

Evaluation Report

Evaluation of the jetson-customize-camera skill before publication through NVSkills-Eval.

This benchmark summarizes 3-Tier Evaluation from NVSkills-Eval results for the skill. The goal is to document whether the skill is safe, discoverable, effective, and useful for agents before it is published for broader workflow use.

Evaluation Summary

  • Skill: jetson-customize-camera

  • Evaluation date: 2026-06-16

  • NVSkills-Eval profile: external

  • Environment: astra-sandbox

  • Dataset: 2 evaluation tasks

  • Attempts per task: 1

  • Pass threshold: 50%

  • Overall verdict: PASS

Agents Used

  • claude-code

  • codex

Metrics Used

Reported benchmark dimensions:

  • Security: checks whether skill-assisted execution avoids unsafe behavior such as secret leakage, destructive commands, or unauthorized access.

  • Correctness: checks whether the agent follows the expected workflow and produces the correct final output.

  • Discoverability: checks whether the agent loads the skill when relevant and avoids using it when irrelevant.

  • Effectiveness: checks whether the agent performs measurably better with the skill than without it.

  • Efficiency: checks whether the agent uses fewer tokens and avoids redundant work.

Underlying evaluation signals used in this run:

  • security (Security): checks for unsafe operations, secret leakage, and unauthorized access.

  • skill_execution (Skill Execution): verifies that the agent loaded the expected skill and workflow.

  • skill_efficiency (Efficiency): checks routing quality, decoy avoidance, and redundant tool usage.

  • accuracy (Accuracy): grades final-answer correctness against the reference answer.

  • goal_accuracy (Goal Accuracy): checks whether the overall user task completed successfully.

  • behavior_check (Behavior Check): verifies expected behavior steps, including safety expectations.

  • token_efficiency (Token Efficiency): compares token usage with and without the skill.

Test Tasks

The benchmark dataset contained 2 evaluation tasks:

  • Positive tasks: 2 tasks where the skill was expected to activate.

  • Negative tasks: 0 tasks where no skill was expected.

  • Unlabeled tasks: 0 tasks where positive/negative intent could not be inferred.

Task composition is derived from the evaluation dataset when possible. Entries with expected_skill set are treated as positive skill-activation cases, while entries with expected_skill: null are treated as negative activation cases.

Results

DimensionNumclaude-codecodex
Security2100% (+0%)100% (+0%)
Correctness255% (+40%)91% (+33%)
Discoverability244% (+44%)84% (+33%)
Effectiveness230% (+13%)67% (+48%)
Efficiency249% (+22%)80% (+40%)

Score values show skill-assisted performance. Values in parentheses show uplift versus the no-skill baseline when baseline data is available.

Tier 1: Static Validation Summary

Tier 1 validation passed with observations. NVSkills-Eval ran 1 checks and found 3 total findings.

Top findings:

  • MEDIUM SCHEMA/body_recommended_section: Missing recommended section: '## Instructions' (skills/jetson-customize-camera/SKILL.md)

  • MEDIUM SCHEMA/body_recommended_section: Missing recommended section: '## Examples' (skills/jetson-customize-camera/SKILL.md)

  • LOW SCHEMA/author_format: Author must be of the form 'Name <email@host>' (skills/jetson-customize-camera/SKILL.md)

Tier 2: Deduplication Summary

This tier was not run or did not produce findings in this report.

Publication Recommendation

The skill is suitable to proceed toward NVSkills-Eval publication based on this benchmark. Skill owners should keep this file with the skill and refresh it when the evaluation dataset, skill behavior, or target agents materially change.

Requirements and compatibility

Written for: Claude Code

Works with: Claude Code, Claude apps, OpenAI Codex, GitHub Copilot, Cursor, Gemini CLI, Kiro, OpenCode, Goose, Amp, Roo Code, JetBrains Junie, Factory Droid, Devin (ex-Windsurf), Mistral Vibe, OpenHands, Letta

“Works with” means the tool reads the open Agent Skills format; a skill written for one tool may use features others lack.

Needs: Names Claude Code tools

Security

What our scanner noticed· Medium risk

  • Large base64-encoded blobs

A static scan is not a guarantee. Skills are third-party content: review a skill before you let an agent use it.

Attribution and licence

Created by
Jetson Team
Publisher
NVIDIA (Official)
Distributed by
NVIDIA
Source
NVIDIA/skills · skills/jetson-customize-camera · commit fce3cca
Licence
Apache License 2.0 (redistribution allowed)
  • LICENSE (added by AutoGineer from LICENSE-APACHE)
  • ATTRIBUTION.md (written by AutoGineer)
Synced

AutoGineer is not affiliated with or endorsed by Jetson Team or NVIDIA.

Download ZIP

The ZIP holds the skill’s files, its licence files and an ATTRIBUTION.md. Download jetson-customize-camera.zip