微信公众号自动化工作流 (WeChat Official Account Automation Workflow)
本文档概述了自动化微信公众号内容创作的 5 个 Skill 的顺序执行流程。
工作流伪代码 (Workflow Pseudocode)
def run_wechat_automation_workflow(initial_input):
# 加载配置
config = load_json("config.json")
# 1. 选题 (Topic Selection)
# 输入: 当前趋势, 细分领域
# 输出: 选定的主题、切入角度、目标读者、拟定标题、关键词
topic_output = execute_skill(
name="Topic",
input={
"niche": initial_input.niche,
"trends": initial_input.trends
},
config=config
)
# 2. 研究 (Research)
# 输入: 选定的主题、关键词
# 输出: 研究笔记、核心洞察、关键事实、来源、Twitter情报
# 工具: 需要网页搜索、浏览器访问、Twitter猎取
research_output = execute_skill(
name="Research",
tools=["web_search", "browser", "read_url", "twitter_hunter"], # 新增 twitter_hunter
input={
"topic": topic_output.selected_topic,
"keywords": topic_output.keywords,
"twitter_config": config.twitter # ← 传入 Twitter 配置
},
config=config
)
# 3. 结构化 (Structure)
# 输入: 研究数据、目标读者、切入角度、语气
# 输出: 文章大纲（含 hook、outline、closing）
structure_output = execute_skill(
name="Structure",
input={
"research_data": {
"key_insights": research_output.key_insights,
"notes": research_output.notes,
"facts": research_output.facts
},
"target_audience": topic_output.target_audience, # ← 从 Topic 传递
"angle": topic_output.angle, # ← 从 Topic 传递
"tone": config.account_tone
},
config=config
)
# 4. 写作 (Write)
# 输入: 大纲、研究数据、目标读者、切入角度、开篇钩子、结尾设计
# 输出: 完整初稿、实际字数、可读性评分
write_output = execute_skill(
name="Write",
input={
"outline": structure_output.outline,
"hook": structure_output.hook, # ← 从 Structure 传递
"closing": structure_output.closing, # ← 从 Structure 传递
"research_data": research_output.notes,
"target_audience": topic_output.target_audience, # ← 从 Topic 传递
"angle": topic_output.angle, # ← 从 Topic 传递
"length_constraints": config.article_length,
"banned_words": config.banned_words
},
config=config
)
# 5. 封装 (Package)
# 输入: 初稿、拟定标题、目标读者、语气
# 输出: 最终标题、摘要、封面图提示词、嵌入图片的文章、SEO关键词
final_package = execute_skill(
name="Package",
input={
"draft": write_output.draft,
"potential_titles": topic_output.potential_titles, # ← 从 Topic 传递
"target_audience": topic_output.target_audience, # ← 从 Topic 传递
"tone": config.account_tone
},
config=config
)
# 6. 发布 (Publish) - 推送到微信
# 输入: 最终标题、摘要、完整文章
# 输出: 推送状态、推送时间、消息ID
if config.push.enabled:
publish_output = execute_skill(
name="Publish",
input={
"title": final_package.title,
"summary": final_package.summary,
"draft_with_images": final_package.draft_with_images,
"pushplus_token": config.push.pushplus_token
},
config=config
)
return final_package, publish_output
数据流图 (Data Flow)
┌─────────┐
│ Topic │
└────┬────┘
│ selected_topic, keywords
│ target_audience, angle, potential_titles
▼
┌──────────┐
│ Research │
└────┬─────┘
│ key_insights, notes, facts, references
▼
┌───────────┐ ← target_audience, angle (from Topic)
│ Structure │
└────┬──────┘
│ hook, outline, closing
▼
┌─────────┐ ← target_audience, angle (from Topic)
│ Write │ ← research_data (from Research)
└────┬────┘
│ draft, actual_word_count
▼
┌─────────┐ ← potential_titles, target_audience (from Topic)
│ Package │
└────┬────┘
│ title, summary, draft_with_images
▼
┌─────────┐ ← pushplus_token (from config)
│ Publish │
└────┬────┘
│ push_status, push_time, message_id
▼
[推送完成]
数据传递原则 (Data Passing Principles)
- 完整传递: Topic 的输出（target_audience, angle, potential_titles）需要贯穿整个工作流
- 层级聚合: Research 的输出在 Structure 和 Write 阶段均被使用
- 配置注入:
config.json中的 tone、banned_words、article_length、push 注入到相关 Skill - 质量联检: 每个 Skill 都有质量自检机制，确保输出符合预期
- 可选推送: Publish Skill 仅在
config.push.enabled = true时执行