/jetson-video-pipeline

name: jetson-video-pipeline
license: "Apache-2.0"
description: >-
  Use when executing and verifying Jetson Video Codec SDK or PyNvVideoCodec
  encode/decode, transcode, segmentation, container decode, AV1, or acceptance
  workflows with exact artifact handoffs.
metadata:
  author: "Vinit Bansal <vinitkumarb@nvidia.com>"
  tags: [jetson, video-codec-sdk, pynvvideocodec, pipeline, nvenc, nvdec]
  languages: [python]
  data-classification: public

Jetson Video Pipeline

Purpose

Execute official-sample codec stages and prove that every consumer used the exact artifact produced by the preceding stage. Use this skill for encode-then-decode verification, native H.264-to-HEVC transcode, PyNvVideoCodec segments, container decode triage, AV1 operation verification, or a compact customer acceptance package.

Prerequisites

• Run execution on the target Jetson with direct GPU access. A fresh validated

schema-1.2 `nvcodec-environment` identity from `jetson-video-setup` is optional. When supplied it is authoritative, and invalid or stale evidence fails closed without local fallback. The agent may obtain it from setup's public read-only probe; it need not be supplied in the customer's prompt.

• Without setup evidence, authenticate only the selected installed surface.

Native routes inspect the fixed dpkg package, package-owned official sample sources, build tools, and non-stub linkage. PyNvVideoCodec routes require an authenticated setup environment or the caller's exact absolute `pynvc_interpreter`; never scan for a venv. Before asking for that path, invoke setup's public probe when that skill is installed and inspect its typed result. These read-only checks install, repair, register, and smoke-test nothing.

• Recipe-bearing routes require sibling `jetson-video-recipe` and one of its

validated schema-2 recipes. If its canonical public CLI is present, invoke it; if absent, preserve `dependency_required`, name that skill, and tell the user to install it and retry the stage. Recipe-free decode/segmentation routes do not acquire that dependency.

• `capability_report` is an optional encode-request member, never a required

one. The established authority for PyNvVideoCodec encoder capabilities is the `capabilities` block of the schema-1.2 `nvcodec-environment` artifact; an encode request that omits `capability_report` is fully supported and reads that block. When capability-owned freshness is wanted, a request may additionally carry a schema-1.0 `nvcodec-encoder-capability-report` produced by `jetson-video-capability` from the *same* environment artifact. When supplied, that report becomes the selected Py encoder API evidence for the check; it does not replace the environment artifact or its readiness facts. Do not add the member to an independently constructed request merely because the `pynvc` surface may be selected. It is optional on either surface, affects Py capability classification only, and should be omitted for native; absence never fails.

• When setup is installed, read its shared

[video content policy](../jetson-video-setup/references/video-content.md) and apply its input gate before any normal pipeline dry run or execution. Setup is not required solely for this policy: without it, require one exact user-selected path or URL, never substitute catalog or synthetic media, and preserve source URL, license, attribution, path, size, and SHA-256.

• PyNvVideoCodec routes that encode or invoke `advanced/decode.py`, including

encode/decode, segmentation, and Py container triage, require a separately validated `full-samples` venv. The default `pynvc-smoke` environment is a setup-readiness proof and must block these routes before workspace creation; return a structured `jetson-video-setup` dependency and provision a new full-samples venv rather than upgrading it in place. If that skill is absent, tell the user to install it before retrying.

• The direct `encode` controller has one narrower consumer exception:

`jetson-video-capability` may bind setup's deterministic one-frame raw fixture for an exact bounded capability smoke operation. That result is operation evidence only, never representative pipeline or performance proof.

Compose requested sibling stages

Recipe-free decode and segmentation routes require no sibling when the selected SDK prerequisites already exist. Add `jetson-video-recipe` only for a recipe-bearing stage, `jetson-video-benchmark` only for requested performance, `jetson-video-setup` only for installation, repair, or one read-only handoff when registered Python authority is required, and `jetson-video-capability` only for a requested support verdict or fresh acceptance capability artifact. Use the agent runtime's installed-skill catalog before each stage; do not scan arbitrary directories. If the sibling is present, read its `SKILL.md` and invoke its documented public entry point; pass artifacts as data and never import sibling code. If it is absent, preserve completed stages and artifacts and say, using the actual names: `I can run <stage>, but it requires <skill>, which is not installed. Install <skill> and retry this stage.` Never promote a partial workflow to complete or require an optional sibling.

Instructions

1. Apply the scope boundary. For a request solely for objective quality metrics, including PSNR or SSIM, state only that this skill does not provide them and that a separately authorized quality workflow is required, then stop. Do not name or recommend an external tool, and do not offer to configure or run the comparison. For a request limited to capture, transport, AI, display, or glass-to-glass latency, state that those stages are outside this codec skill and stop without naming, recommending, or offering another tool or workflow. Otherwise proceed immediately to the media gate in step 2; choose `encode_decode`, `native_transcode`, `pynvc_se