3.3 KiB

Raw Permalink Blame History

S-18 Slide Prompt V1

Status

Prompt saved before generation.
Line: production-light
Review mode: none
Output image: slides/s18/s18-v1-image.png

Based On Brief

slides/s18/s18-v1-brief.md

Reference Inputs

Image 1: projects/2026-07-01-为什么是我们/slides/s01/s01-v1-image.png as style continuity reference, not edit target.
Image 2: projects/2026-07-01-为什么是我们/slides/s10/s10-v1-image.png as style continuity reference, not edit target.
Image 3: projects/2026-06-29-搭上AI这条船/slides/s01/s01-v1-image.png as style continuity reference, not edit target.
Image 4: projects/2026-06-29-搭上AI这条船/slides/s10/s10-v1-image.png as style continuity reference, not edit target.

Operation Notes

Use the loaded reference images only for visual continuity.
Do not call Garden generation scripts.
Backup fact-check source was not opened or used; no factual claim needed verification.

Advisor / Host-Native Record

Skill mode command: node C:\Users\wangq\.agents\skills\gpt-image-2\scripts\check-mode.js --json
Mode result: B-or-C; recommendation: host-or-advisor; Garden disabled; API key absent; effective path: Mode B because host image_gen is available.
Skill/template files read: SKILL.md, prompt-writing.md, visual-report-page.md, dense-explainer-slides.md, educational-diagram-slide.md, and system imagegen/SKILL.md.
Project files actually read: AGENTS.md, project.md, execution-plan.md, slides/slides.md, visual-system/visual-system.md, style-prompt-master.md, motif-registry.md, and the three accepted intake files.

Final Prompt

Create a 16:9 full-page visual slide image for an internal sales-alignment PowerPoint. Use Image 1-4 as style continuity references only: dark navy-black executive AI maritime operation map, subtle coordinate grid, radar/compass overlays, cyan signal paths, restrained gold highlights, serious deepwater business-system tone.

Page: P18 / S-18. Main title in large readable Chinese: "我们不只追求能跑，还追求好改". Core judgment: "AI 出错后，要能被发现、解释、纠正和沉淀".

Create a central calibration loop, like a reliable expert-agent control system rather than a generic dashboard. Loop nodes in order: "AI 输出" → "专家审查" → "用户反馈" → "错误解释" → "修正记录" → "规则校准" → "再次输出". Use cyan loop lines and restrained gold on the calibration nodes.

Around the loop place five capability badges: "可发现", "可解释", "可纠正", "可沉淀", "可迭代". Add subtle visual details: feedback slot, calibration knob, version iteration curve, captured error marker entering a structured data layer.

Bottom judgment bar: "交付纠错机制和校准机制". If space allows, include smaller text: "不是只交付能跑的结果".

Keep the page sober, technical, and reassuring. Do not imply AI never fails. Avoid generic operations dashboard, stock office people, human portraits, robots, logos, watermarks, long paragraphs, and bright cyberpunk neon overload.

Negative / Avoid

Do not promise zero errors. Do not make a generic monitoring dashboard. Do not use long failure text.

Output Intent

Copy host-generated image to projects/2026-07-01-为什么是我们/slides/s18/s18-v1-image.png.

3.3 KiB Raw Permalink Blame History