| name | ip-cognition-script-polisher |
| description | Polish rough topics, short opinions, or draft scripts into 30-60 second IP incubation, business cognition, and AI cognition short-video voiceover packages for whiteboard infographic videos. Use when Codex needs to preserve a user's core stance while producing polished_voiceover.md, voiceover_segments.json, and visual_beats.json with hook/反常识/例子/转折/方法/结论 segments, spokenAnchors, visualIntent, and duration-checkable output. |
IP Cognition Script Polisher
Overview
Use this skill to turn a topic, rough opinion, or partial draft into a reviewable script package for IP incubation, business cognition, and AI cognition short videos. The output is not a final video; it is the upstream script contract for whiteboard infographic planning, voiceover timing, and later rendering.
Workflow
- Read the user's brief and extract the core stance, target audience, promised value, and any non-negotiable wording.
- Preserve the user's core viewpoint. Sharpen framing and rhythm, but do not reverse the stance or smuggle in a different conclusion.
- If the input is only a topic, infer a conservative thesis and make the inference explicit in
polished_voiceover.md under Stance basis.
- Produce exactly six voiceover segments in this role order:
hook, 反常识, 例子, 转折, 方法, 结论.
- Keep the total estimated voiceover length between 30 and 60 seconds. Default target is 45 seconds unless the user specifies otherwise.
- Add
spokenAnchors and visualIntent to every segment. Use anchors that appear verbatim in the segment text or caption.
- Create only necessary visual beats. Do not turn every sentence into a board object; group the script into 3-6 strong visual ideas.
- Validate the package with the bundled script before handing it off.
Outputs
Write these files into the requested package directory, usually script/ inside a workflow project:
script/
├── polished_voiceover.md
├── voiceover_segments.json
└── visual_beats.json
voiceover_segments.json is the source of truth for narration. visual_beats.json is the source of truth for the next infographic-planning skill. polished_voiceover.md is the human review surface.
Segment Rules
hook: Start with a sharp, memorable tension. Avoid vague openings such as "今天聊聊".反常识: State the counterintuitive claim without overclaiming beyond the user's brief.例子: Use one concrete scene, comparison, or mini case. Prefer business/IP/AI operator details over generic life advice.转折: Reframe the issue and connect the example back to the core thesis.方法: Give a repeatable 2-4 step method, checklist, or decision rule.结论: End with a compact sentence that can become the final subtitle or board title.
Each segment must include:
{
"id": "seg-01-hook",
"role": "hook",
"text": "口播正文",
"caption": "字幕文本",
"visualIntent": "这一段应该让白板画面表达什么",
"spokenAnchors": ["必须出现在 text 或 caption 里的短语"]
}
Optional fields such as boardId, targetElement, targetDurationSec, and pauseAfter are useful for downstream timing and board control.
References
Read references/script_patterns.md when shaping a script from a very short topic or when the structure feels flat. Read references/schema.md before changing JSON fields or building downstream consumers.
Validation
Run:
python3 scripts/validate_script_package.py --package-dir ../../_examples/polisher/output
For an external project package:
python3 /path/to/ip-cognition-script-polisher/scripts/validate_script_package.py --package-dir /path/to/project-output/script
Validation checks required fields, exact role coverage/order, spokenAnchors, matching visual beat segment IDs, and estimated duration.
