release-showcase-manager

Skill: release-showcase-manager

Overview

This skill is specifically used to manage large-scale showcase projects during AI model releases. It is not merely a video editing tool, but anend-to-end engineering management framework. It guides the Agent to start from model capability research, design and implement specific development tasks (such as Web generation, code engineering, mobile applications, etc.), and synchronously complete recording, note-taking, and performance evaluation during the development process, ultimately producing showcase materials that can be used for release.

This skill is deeply integrated with doc-todo-log-loop, ensuring that in complex long-term tasks it can resume from any breakpoint at any time and maintain context consistency.

---

Managed Workspace Structure

When using this skill to manage a project, the project's workspace should have the following structure (refer to assets/template_workspace/）：

• notes/: Unified knowledge and records directory. Includes model documentation, Research, development logs, delivery logs, etc.
• showcases/: Core output directory, each subdirectory represents an independent showcase project (such as web-app-gen/, mobile-ocr/）。
• video-raw/: Stores original screen recording materials.
• video-clipped/: Stores finished videos after preliminary trimming and processing.

Document Naming & Classification Standards (Naming & Classification): Under the notes/ directory, all documents must follow the mandatory naming format:yyyy-mm-dd-hh 类型 名字.md。

• Type Definitions：
  • 笔记: Personal essays, scattered ideas, or scattered notes submitted by users.
  • 文档摘录: Documents or reference materials copied or excerpted from external sources.
  • Research: Online research results for a specific topic.
  • 开发日志: Development process records and chain-of-thought analysis for Showcase or demonstration projects.
  • 交付日志: Records delivery package contents, snapshots, and update notes.
• Language Requirements: Document content and naming must use Chinese.

Structural Consistency Constraints (Structural Consistency)： showcases/, video-raw/ and video-clipped/ mustmaintain a completely consistent internal subdirectory hierarchy.

• Use showcases/ as the baseline: Users usually store all original materials (screen recordings, screenshots) directly during development in showcases/ under the corresponding project subdirectory.
• Agent Responsibilities: The Agent should use showcases/ as the baseline source, be responsible for syncing/moving original materials to video-raw/, and store processed videos in video-clipped/, while strictly maintaining a consistent directory hierarchy. Directly piling files in the root directory is prohibited.

The workspace must be configured with Git LFS to manage large video materials.

---

Core Workflow (The Pipeline / Sub-commands)

This skill supports phased execution through the following subcommands (Workflows). Users can request at any time: "Enter phase N" or "Execute [subcommand name]".

1. [discovery] Discovery and Collection: See workflows/01-discovery-and-collection.md. Collect model profiles and materials.
2. [planning] Solution Design: See workflows/02-planning-and-design.md. Initialize doc-todo-log-loop and design Scenario.
3. [execution] Development Implementation: See workflows/03-development-and-execution.md. In showcases/ for the encoding implementation.
4. [capture] Recording and documentation: See workflows/04-capture-and-notetaking.md. Record the video and synchronously update notes/。
5. [evaluation] Evaluation and insights: See workflows/05-evaluation-and-insights.md. Produce a formal model evaluation report.
6. [finalization] Final output: See workflows/06-finalization.md. Produce the finished video and archive it.
7. [packaging] Delivery package packaging: See workflows/07-handover-packaging.md. Package the asset bundle intended for video editors.

---

Dependencies and skills

• doc-todo-log-loop: Used for task tracking and state preservation throughout the process.
• showcase-video-processor: Used for fine-tuning the video in post-production.
• oneshot-website / ppt-skill: Used as an efficient output tool during the development phase.

---

Best practices

• Continuous logging: Each time a development phase is completed, you must use doc-todo-log-loop to update the log.
• LFS awareness: Before committing video files, ensure that .gitattributes。
• Multidimensional recording: In the notes, record not only "what was done," but also where the model showed "surprise" or "struggle."
• Persisting asset trimming instructions: Record all asset trimming instructions provided by the user (timestamps, target duration, speed ratios, etc.) completely in video-clipped/CLIPPING_LOG.md to ensure that every edited asset clip is documented and traceable.
• Simplifying delivered assets: When packaging the delivery package, proactively remove lengthy date prefixes and debug suffixes. Keep directory names in a concise {model} {id} {name} format.
• Flexibility of the storyboard script: In the delivery document, clearly inform the editor that the storyboard script is a suggestion based on "logical points," and encourage secondary creation based on visual rhythm while ensuring a closed narrative loop.
• Delivery traceability and update notes: Under the notes/ directory, maintain yyyy-mm-dd-hh 交付日志 {名字}.md to record a snapshot of the directory structure for each output. If the asset bundle is updated, use overwrite-style replacement accompanied by UPDATE_NOTES.md to explain the changes.