360 Secure SkillHub
360 安全技能中心

媒体交付包校验

该 Skill 使用 Python 校验脚本和 ffprobe,按用户提供的显式契约检查媒体交付包的文件成员、视频编码/帧率/尺寸/音频、图片尺寸及 SRT 字幕时序;只输出检查结果,不修改、复制、发布或提升文件,也不判断画面可读性和编辑质量。

分类:数据
xxjrq
xxjrq
热度 2.39
最近更新 2026-10-27 21:27:40

安装

发给 Agent
帮我安装这个技能:npx secure-skills-cli add xxjrq/selfmedia-visual-qa-release
兼容性
通用兼容
联网要求
不联网
数据出网
数据不出本机
Metadata
name
selfmedia-visual-qa-release
description
Validate a rendered media delivery package against an explicit file and media contract, including video streams, image sizes, and subtitle timing; report failures without promoting or overwriting files.
license
MIT-0
metadata
{"version":"1.0.0"}

Media Release Package QA

Use this skill to inspect a finished media package before handoff. First identify the user's actual contract: required filenames, video codec/frame rate/dimensions/audio requirements, image dimensions, subtitle rules, and destination. Do not treat the included Douyin five-file profile as a universal platform requirement.

Run the validator with an explicit package directory and contract:

python3 scripts/validate_release_package.py /path/to/package --contract contracts/douyin-five-file.json

The included profile is a reusable example based on a project-specific five-file handoff: copy, video, 3:4 cover, 4:3 cover, and SRT captions. Its dimensions are declared in the contract. Adjust the JSON contract when the actual user or platform requirements differ; do not silently add requirements based on convention.

The validator requires ffprobe on PATH. It checks exact file membership, probe-readable media, declared codec/frame rate and image dimensions, and valid ordered SRT cue times within video duration. It reports findings only: it never edits, deletes, copies, or promotes the package. On any failure, preserve the user's package and report the precise field. A pass is a check against the selected local contract, not proof of platform acceptance or visual quality.

Run local regression checks with python3 scripts/self_test.py. Review images at intended display size and inspect the video visually; automated metadata checks cannot judge legibility, crop, visual artifacts, or editorial quality.