spring-ai-movie-poster-copywriter

name	spring-ai-movie-poster-copywriter
description	Build a movie-poster one-line recommendation pipeline that takes movie title and description, gathers context from Douban and encyclopedia pages, and generates 3-5 Chinese recommendation lines with exactly 12 Chinese characters each. Use when Codex needs to implement crawler-to-prompt workflows, output validation, and recommendation generation APIs.

Spring AI Movie Poster Copywriter

Workflow

1. Collect request payload:

• Require movieTitle and movieDescription.
• Accept optional year, genre, and tone.

2. Retrieve context:

• Query Douban and encyclopedia pages using search first, then fetch detail pages.
• Extract only concise facts: genre, setting, themes, cast keywords, awards, and plot hooks.
• Ignore user-generated long comments and spoiler-heavy content.

3. Normalize context:

• Convert raw HTML into a compact context block under 1200 Chinese characters.
• Keep only deduplicated facts and keywords.
• Store retrieval source URLs for traceability.

4. Generate candidates:

• Load prompt templates from references/prompt-template.md.
• Ask model to generate 8-10 candidates first.
• Keep temperature moderate (0.6 to 0.8) to balance creativity and control.

5. Validate and filter:

• Run scripts/validate_slogans.py against candidate lines.
• Keep final 3-5 lines that satisfy length and uniqueness checks.
• If fewer than 3 survive, regenerate with stricter prompt constraints.