wantsong
/
video-workbench
1
0
Fork 0
Code Issues Pull Requests Packages Projects Releases Wiki Activity
video-workbench/docs/ccpe-system-handoff.md

4.3 KiB
Raw Blame History

CCPE System Handoff For Video Workbench

This document is the handoff for future work in:

C:\Users\wangq\Documents\Codex\ccpe-system

It records the CCPE assets needed to support Video Workbench.

Core Principle

Use scenario-first layering.

Do not force every old prompt into Skill, Agent, and Runtime layers. Preserve a working Lite kernel when it is the correct production form. Promote to Agent Spec, Skill, Model Card, or Runtime only when the video workflow needs that layer.

Existing Actor To Upgrade

Previous single-agent actor:

强哥的徒弟

Likely first target:

agents/lite/qiangge-apprentice.prompt.md

Purpose:

Transform deep essays and models into public-facing short video plans, scripts, visual prompts, and production packages.

This actor should remain a CCPE-owned local work asset. Video Workbench may request, consume, and validate it, but should not store its canonical prompt or agent definition.

Recommended CCPE Build Sequence

  1. Upgrade 强哥的徒弟 into a verified CCPE-Lite artifact.
  2. Validate it through a real Video Workbench project when the user starts one.
  3. Preserve regression tests against the original workflow:
    • angle proposals
    • video outline
    • storyboard logic
    • shot deep production
  4. Run Scenario Probe before promoting to Agent Spec or Skill.
  5. Extract stable embedded models into Model Cards only when they are reusable beyond this actor.
  6. Build small Skills only for repeatable actions:
    • source-to-episode splitter
    • angle proposal generator
    • global visual style builder
    • storyboard logic builder
    • shot deep-production generator
    • image prompt generator
    • voiceover script generator
    • publish copy generator
  7. Build a Short Video Production Runtime only after the core stages are stable.
  8. Validate the runtime with a trial Video Workbench project.

Candidate Asset Map

Asset Candidate Type Purpose
Qiangge's Apprentice CCPE-Lite / Agent Spec Primary dimensional output actor
Dimensional Output Model Model Card Extract deep model into public-facing point
Series Architect Agent / Skill Split long essay/model into episodes
Angle Proposal Generator Skill Produce three distinct entry angles
Tone And Outline Builder Skill Produce video outline and global style
Storyboard Logic Builder Skill Produce shot logic with timing and word budget
Shot Deep Production Skill Produce visual plan, image prompt, voiceover, design notes
Image Prompt Generator Skill Generate consistent image prompts
Voiceover Script Generator Skill Fit narration to duration budget
Short Video Production Runtime Runtime Coordinate full video package production

Runtime Boundary

The first runtime should stop at a production package:

  • selected angle
  • video outline
  • storyboard logic
  • shot deep-production files
  • image prompts
  • voiceover script
  • optional image/audio assets
  • editing notes
  • publishing copy

Do not automate manual editing or platform publishing in the first version.

Relationship To Writing Runtime

Article production and video production should remain separate runtimes.

They share sources and sometimes models, but they differ in:

  • target output
  • audience
  • timing constraints
  • visual constraints
  • asset generation
  • evaluation criteria

Relationship To Deployable Systems

CCPE can design local work agents and can also specify production/business agents for deployable systems. For deployable systems, CCPE owns specifications, boundaries, authority, evaluation, and invocation contracts.

Implementation that depends on LangGraph, CrewAI, server deployment, persistence, monitoring, or product runtime constraints belongs in the target software project, such as a future work-projects repository. Video Workbench remains the local workflow and production-record workspace.

Validation Target

The first validation project should use one existing deep essay and one existing model document from Knowledge Vault.

Expected success:

  • a clear episode or standalone scope
  • three genuinely different angles
  • a selected angle trace
  • a coherent visual/narration protocol
  • a complete storyboard
  • shot-level prompts and voiceover
  • a package ready for manual editing