hackathon-doc-writer（黑客马拉松文档作家）

hackathon-doc-writer

Goal

Generate structured technical documentation artifacts (ADR, PRD, feature specs) for a hackathon project using the appropriate template.

Trigger Conditions

Use this skill when:

• MVP scope is locked and needs to be documented in a PRD
• An architectural decision has been made that should be recorded in an ADR
• A specific feature needs a formal spec before implementation begins
• Invoked during Phase 4 (Project Planning), after hackathon-scope-cutter completes
• Can be re-invoked any time a new architectural decision is made during implementation

Inputs

Outputs

Rules

1. Use the corresponding template from templates/ directory as the output structure.
2. Do not omit any section from the template; use [TBD] for missing information.
3. Keep language direct and scannable — no filler paragraphs.
4. For ADR: capture exactly one architectural decision per document.
5. For PRD: include success metrics even if they are estimates.
6. For feature-spec: include acceptance criteria as testable conditions.
7. Flag all [TBD] sections in missing_fields output.

Output Format

document_type: "<ADR|PRD|feature-spec>"

missing_fields:
  - "<section name>"

document: |
  <Full Markdown document content>

Example

Input:

document_type: "ADR"
project_title: "AnchorAI"
problem_statement: "How might we provide always-available emotional support for college students?"
mvp_features:
  - "GPT-4 emotional check-in conversation"
  - "Session memory"
tech_stack: ["Python", "FastAPI", "React", "OpenAI API", "Redis"]
architecture_decisions:
  - title: "Session memory storage"
    context: "Need to persist conversation summaries between sessions"
    options: ["Redis (in-memory)", "SQLite (file)", "OpenAI conversation history API"]
    decision: "Redis"
    rationale: "Fastest to set up; sufficient for demo; team familiar with it"

Output:

document_type: "ADR"
missing_fields: []
document: |
  # ADR-001: Session Memory Storage

  **Date:** 2025-06-14
  **Status:** Accepted

  ## Context
  AnchorAI must recall emotional context from past user sessions to deliver its core wow factor.
  Memory must persist across browser refreshes for the demo.

  ## Decision
  Use Redis as an in-process session store keyed by a hardcoded demo user ID.

  ## Consequences
  - Positive: Setup takes <30 minutes; team knows Redis well
  - Negative: Data is lost on server restart; not production-ready
  - Risk: Redis unavailable on demo machine → mitigation: fallback to in-memory dict

Context Files

Knowledge Base

• knowledge/hackathon-reference-architecture.md
• knowledge/hackathon-mvp-strategy.md

Templates

• templates/ADR-template.md
• templates/PRD-template.md
• templates/feature-spec-template.md

Playbooks

• playbooks/hackathon-workflow.md