hackathon-code-implementer

hackathon-code-implementer

目标

为黑客马拉松项目任务提供结构化的实现指导，包括适合快速制作原型的代码模式、集成策略和捷径。

---

触发条件

在以下情况下使用此技能：

• 即将实现 hackathon-task-planner 中的特定任务
• 需要实现指导、代码框架或捷径模式
• 开发者受阻，需要一个具体的起点
• 需要评估任务预算是否有超出预算的 [HIGH-RISK] 风险
• 每个任务调用一次；遍历 hackathon-task-planner 中的完整任务列表

---

输入

输入	类型	必填	描述
task_title	字符串	是	正在实现的任务（来自 hackathon-task-planner）
task_description	字符串	是	任务必须实现的详细说明
tech_stack	字符串[]	是	使用的技术
mvp_demo_flow	对象[]	是	来自 hackathon-scope-cutter 的演示流程步骤
time_budget_hours	数字	是	分配给此任务的小时数
existing_code_context	字符串	否	相关的现有代码片段或文件结构
fake_ok	布尔值	否	是否接受硬编码或模拟数据（默认值：false）

---

输出

输出	描述
implementation_plan	完成任务的有序子步骤列表
code_scaffolds	可作为起点的关键代码片段、模式或桩代码
integration_points	此任务与其他组件的连接位置
shortcuts	适合黑客马拉松的捷径（模拟、硬编码、库）
gotchas	常见的失败情况及规避方法
done_criteria	表明任务已完成的条件

---

规则

1. 优先考虑可运行的代码，而非整洁的代码；明确指出技术债务。
2. 尽可能推荐现有库，而非自定义实现。
3. 如果 fake_ok 为 true，同时提供模拟或桩代码模式和实际实现。
4. code_scaffolds 应保持精简——仅提供入口点，而非完整实现。
5. done_criteria 必须能在 time_budget_hours 内观察和验证。
6. 对于任何可能耗时超过 time_budget_hours 50% 的子步骤，标记为 [HIGH-RISK]。
7. 不要生成生产级架构；优先确保演示完整。

---

输出格式

implementation_plan:
  - step: <number>
    action: "<what to do>"
    risk: "<[HIGH-RISK]|normal>"

code_scaffolds:
  - label: "<purpose>"
    language: "<language>"
    snippet: |
      <code>

integration_points:
  - component: "<name>"
    connection: "<how this task connects>"

shortcuts:
  - shortcut: "<description>"
    trade_off: "<what is sacrificed>"

gotchas:
  - "<pitfall>"

done_criteria:
  - "<verifiable condition>"

---

示例

输入：

task_title: "Implement session memory read/write (T-03)"
task_description: "Store and retrieve a rolling summary of the user's emotional state across chat sessions using Redis."
tech_stack: ["Python", "FastAPI", "Redis", "OpenAI API"]
mvp_demo_flow:
  - step: 2
    action: "User describes recurring work stress"
    outcome: "AI references last week's similar conversation from memory"
time_budget_hours: 3
fake_ok: false

输出：

implementation_plan:
  - step: 1
    action: "Install redis-py; add Redis client singleton to app startup"
    risk: "normal"
  - step: 2
    action: "After each GPT-4 response, call GPT-4 to generate a 2-sentence session summary and write to Redis key user:{id}:summary"
    risk: "normal"
  - step: 3
    action: "On each new conversation, prepend the stored summary to the system prompt"
    risk: "[HIGH-RISK] — summary injection may push context window near limit; keep summary ≤100 tokens"
  - step: 4
    action: "Test by running two sessions and verifying AI references prior context"
    risk: "normal"

code_scaffolds:
  - label: "Redis memory read/write"
    language: "python"
    snippet: |
      import redis
      r = redis.Redis(host="localhost", port=6379, decode_responses=True)

      def get_memory(user_id: str) -> str:
          return r.get(f"user:{user_id}:summary") or ""

      def save_memory(user_id: str, summary: str):
          r.set(f"user:{user_id}:summary", summary, ex=86400)

integration_points:
  - component: "OpenAI chat endpoint (T-02)"
    connection: "Prepend get_memory() result to system prompt on every request"

shortcuts:
  - shortcut: "Use a single hardcoded user_id='demo' for the hackathon"
    trade_off: "No real multi-user support; acceptable for single-demo session"

gotchas:
  - "Redis not running on demo machine → add a startup health check with a clear error message"
  - "Summary growing too long → hard-cap at 150 tokens before injection"

done_criteria:
  - "Starting a new chat session shows AI referencing content from the previous session"
  - "Redis key exists and contains non-empty summary after first session ends"

---

上下文文件

知识库

• ../hackathon-shared-resources/knowledge/hackathon-reference-architecture.md
• ../hackathon-shared-resources/knowledge/hackathon-tools.md
• ../hackathon-shared-resources/knowledge/hackathon-mvp-strategy.md
• ../hackathon-shared-resources/knowledge/hackathon-common-failures.md

操作手册

• ../hackathon-shared-resources/playbooks/hackathon-workflow.md