← 返回 Skills 市场
kotot

dpp-pipeline

作者 kotot · GitHub ↗ · v1.0.2 · MIT-0
cross-platform ✓ 安全检测通过
79
总下载
0
收藏
0
当前安装
3
版本数
在 OpenClaw 中安装
/install dpp-pipeline
功能描述
Use when you need to turn a local source video and one product image into a single-product placement video with the DPP pipeline.
使用说明 (SKILL.md)

DPP Pipeline

Run the DPP storyboard, placement, and compose-best flow from a local workspace.

The skill bundles its Python runtime under runtime/, but it does not treat the installed skill directory as the project workspace. Inputs and outputs live under the caller's current working directory by default. Override that with DPP_WORKDIR when needed.

When to Use

  • Generate storyboard.json from a local video with dpp-storyboard.
  • Generate a single-product material config from one product image under assets/.
  • Select the best storyboard segment for one product image with dpp-placement.
  • Run the final best-segment composition flow with dpp-compose-best.
  • Replace the best segment in the original video with the generated clip using dpp-final-cut.

Do not use this skill for multi-product batch processing. This skill handles one local video and one product image at a time.

Dependencies

  • Python 3.11+
  • ffmpeg
  • Network access to Ark and TOS
  • Ark environment variables in the target workspace .env
  • TOS_BUCKET, TOS_AK, TOS_SK, TOS_ENDPOINT, and TOS_REGION when compose-best must auto-upload the reference clip or image
  • TOS_OBJECT_PREFIX, TOS_ENABLE_HTTPS, and TOS_FORCE_ENDPOINT are optional TOS tuning variables

Install Python dependencies by running:

scripts/bootstrap_runtime.sh

Workspace Layout

The runtime lives inside the installed skill. The caller workspace holds inputs, configs, logs, and outputs:

\x3Cworkspace>/
  .env
  assets/
    \x3Csingle product image>
  video/
    demo.mp4
  configs/
    placement_material.json
  log/
  output/
    \x3Cvideo_stem>/
      finalCut/

Initialize a new workspace with:

scripts/init_workspace.sh

That command creates assets/, video/, configs/, and output/, and copies sample files into place.

Workflow

  1. Run scripts/bootstrap_runtime.sh.
  2. If the target workspace is empty, run scripts/init_workspace.sh.
  3. Ensure the workspace .env contains Ark credentials and defaults.
  4. Ensure video/demo.mp4 exists before storyboard, unless the user will pass --video.
  5. Ensure assets/ contains exactly one product image before material generation, unless the user will pass --image.
  6. Run scripts/run_storyboard.sh.
  7. Run scripts/run_generate_material.sh.
  8. Run scripts/run_placement.sh.
  9. Run scripts/run_compose_best.sh.
  10. Run scripts/run_final_cut.sh.

Do not skip stage ordering unless the user explicitly asks to reuse an existing artifact.

Review Pauses

Do not pause by default.

  • If the user wants review pauses, stop after each requested stage and report the artifact path.
  • If the user asks to continue, move directly to the next stage when the required inputs already exist.
  • If the user asks to rerun a later stage, reuse earlier artifacts unless the input changed.

Typical checkpoints:

  • After storyboard: review segmentation quality and thumbnail coverage.
  • After placement: review best_segment_index, ranking, and placement rationale.
  • After compose-best: review composition_result.json, prompt quality, and reference-media URL status.
  • After final-cut: review finalcut_result.json and verify the output video at output/\x3Cvideo_stem>/finalCut/final_cut.mp4.

Execution Rules

  • Run the stage scripts from the caller workspace, or set DPP_WORKDIR to the target workspace path.
  • Keep the installed skill directory immutable except for .venv/ and the generated runtime/ bundle.
  • Always bootstrap the runtime before the first stage command.
  • dpp-storyboard, dpp-placement, and dpp-compose-best can each take more than 10 minutes. Do not assume the process is hung just because it is quiet for a while; warn the user that the stage is long-running and let it complete naturally.
  • After each stage, report the main artifact path.
  • Before compose-best auto-upload, verify TOS_BUCKET, TOS_AK, TOS_SK, TOS_ENDPOINT, and TOS_REGION if the user did not provide --reference-video-url or DPP_REFERENCE_VIDEO_URL.
  • This skill sends prompts and media metadata to Ark. Compose-best may also upload the reference clip and product image to TOS.

Detailed commands live in references/commands.md.

Scripts

  • scripts/bootstrap_runtime.sh
  • scripts/init_workspace.sh
  • scripts/run_storyboard.sh
  • scripts/run_generate_material.sh
  • scripts/run_placement.sh
  • scripts/run_compose_best.sh
  • scripts/run_final_cut.sh

Each script forwards all CLI arguments to the underlying Python module.

安全使用建议
This skill appears to do what it claims: it uploads local media to Ark (required) and can optionally upload reference media to a TOS bucket if TOS credentials are provided. Before installing/running: (1) provide only a dedicated workspace directory and avoid placing unrelated secrets in its .env (the skill will load that file); (2) be aware scripts will create a local .venv and pip-install packages from PyPI (network activity during bootstrap); (3) do not run this on sensitive videos you don't want uploaded to Ark/TOS; (4) if you require higher assurance, inspect runtime/src/dpp_storyboard/ark_client.py and compose/upload paths and test in an isolated environment. Overall the required binaries, env vars, and network behavior are proportionate to the stated purpose.
能力标签
requires-sensitive-credentials
能力评估
Purpose & Capability
Name, description, and runtime code all implement a DPP storyboard/placement/compose/final-cut pipeline. Required binaries (python3, ffmpeg), the primary env var (ARK_API_KEY), and optional TOS credentials align with the documented behavior (uploading media to Ark, optionally uploading reference media to TOS).
Instruction Scope
SKILL.md and the stage scripts confine operations to the caller workspace (or DPP_WORKDIR) and describe the exact workflow. The runtime reads a workspace .env for Ark/TOS configuration, uploads media to Ark, and may upload reference media to TOS when requested — these actions are consistent with the stated purpose. There are no instructions to read unrelated system files or exfiltrate data to unknown endpoints.
Install Mechanism
The registry has no formal install spec, but the bundled scripts bootstrap a Python virtualenv and pip-install dependencies from PyPI (runtime/pyproject.toml includes 'tos' and 'volcengine-python-sdk[ark]'). This is expected for a Python-based tool but means network access to PyPI is used during bootstrap and packages will be installed on the host.
Credentials
The only declared required env var is ARK_API_KEY (primary credential) which the code uses. TOS_* variables are read as optional configuration when auto-upload is enabled. The runtime intentionally loads a workspace .env (via DPP_DOTENV_PATH) — users should avoid placing unrelated secrets in that file.
Persistence & Privilege
always is false and the skill does not request persistent platform privileges. Bootstrap creates a local .venv under the skill directory; scripts and code do not modify other skills or system-wide agent settings.
如何使用
  1. 确保已安装 OpenClaw(本地或 Docker 部署)
  2. 在对话框中输入安装命令:/install dpp-pipeline
  3. 安装完成后,直接呼叫该 Skill 的名称或使用 /dpp-pipeline 触发
  4. 根据 Skill 的参数说明提供必要输入,即可获得结构化输出
版本历史
v1.0.2
dpp-pipeline 1.0.2 - Shortened and clarified the skill description for better readability. - No functional or behavioral changes; documentation only.
v1.0.1
dpp-pipeline 1.0.1 - Updated compatibility and documentation to clarify required and optional TOS credentials. - Now lists `TOS_SK` and `TOS_REGION` as required for compose-best auto-upload, with additional optional TOS tuning variables. - Expanded instructions and environment variable requirements for TOS integration. - No code or functionality changes.
v1.0.0
Initial release of dpp-pipeline. - Provides an end-to-end workflow for processing a single local video and product image with DPP: storyboard, material config, placement, composition, and final-cut stages. - Requires Python 3.11+, ffmpeg, Ark and TOS credentials as needed. - Bundles workspace initialization, runtime bootstrapping, and clear execution scripts for each processing stage. - Designed for workspace input/output in the caller's directory (or via DPP_WORKDIR). - Supports user-invocable review checkpoints after each stage. - Not intended for multi-product or batch scenarios.
元数据
Slug dpp-pipeline
版本 1.0.2
许可证 MIT-0
累计安装 0
当前安装数 0
历史版本数 3
常见问题

dpp-pipeline 是什么?

Use when you need to turn a local source video and one product image into a single-product placement video with the DPP pipeline. 它是一个面向 Claude Code / OpenClaw 的 AI Agent Skill 插件,目前累计下载 79 次。

如何安装 dpp-pipeline?

在 OpenClaw 或 Claude Code 对话框中运行命令「/install dpp-pipeline」即可一键安装,无需额外配置。

dpp-pipeline 是免费的吗?

是的,dpp-pipeline 完全免费,采用 MIT-0 许可证,可自由下载、安装和使用。

dpp-pipeline 支持哪些平台?

dpp-pipeline 跨平台运行,可在任意部署了 OpenClaw / Claude Code 的环境中使用(cross-platform)。

谁开发了 dpp-pipeline?

由 kotot(@kotot)开发并维护,当前版本 v1.0.2。

💬 留言讨论