| name | pr-drafting-workflow |
| description | Workflow for creating and managing public relations (PR) drafts for technical projects, incorporating Narrative Heuristic Engineering (NHE) and multi-platform synchronization (GitHub/Notion). |
PR Drafting Workflow
This skill governs the end-to-end process of transforming technical project states into public-facing PR drafts. It prioritizes "narrative" over "list" and ensures high-fidelity synchronization between working drafts, review platforms (GitHub), and long-term knowledge bases (Notion).
Trigger Conditions
- User asks to "prepare PR drafts" or "fill in items for PRs".
- Task involves translating technical milestones/OSS releases into promotional content.
- Multiple projects need parallel draft generation with a consistent structure.
Workflow Steps
1. Information Harvesting
- Cross-Reference Sources: Do not rely on a single source. Synthesize information from:
- GitHub READMEs and technical specs.
- Notion project databases.
- Past press releases (to match tone/style).
- Recent commit history or a "digest" of project progress.
- Gap Analysis: Identify missing items required by the PR template (e.g., specific dates, contact points, "Why now").
2. Narrative Construction (NHE)
- Avoid Bullet-Point Lists: Prefer logical, descriptive prose (Report style).
- Exclude Rhetorical Flourish: For technical PRs, eliminate exaggerated, emotive, or "marketing-heavy" language. Avoid "AI-isms" (e.g., overly dramatic metaphors like "hunting gaps" or "exposing to the swarm"). Do not use phrases that suggest a problem "shakes the foundations" or is a "fundamental challenge." Focus on factual correctness and precise technical definitions.
- Accessibility (General Audience): Use a style that is accessible to non-specialists. When using technical terms, provide a concise explanation in parentheses immediately following the term (e.g., "NTT (Number Theoretic Transform: a mathematical transformation for fast polynomial multiplication)").
- Link Density: Ensure every draft includes a comprehensive "Reference Links" section with direct URLs to GitHub repositories, original papers, demo videos, and official documentation. Do not omit these.
- Focus Areas:
- The "How": Detail the proposed method/technical breakthrough.
- The "Result": Use concrete examples or execution results to prove value.
- Template Adherence: Map the narrative directly to the required items specified by the user/template.
3. Parallel Execution (Scaling)
- For multiple projects, use
delegate_task to spawn subagents.
- Context Injection: Pass the specific project's harvested data and the global NHE guidelines to each subagent to ensure consistency.
- Verification: Read and review subagent outputs before delivery; manually fill gaps for missing projects.
4. Multi-Platform Delivery
- GitHub (Review Loop):
- Post drafts as individual comments on the tracking issue to avoid character limits and allow granular feedback.
- Provide a summary table at the end listing all projects, external verification status, and key dates.
- Notion (Knowledge Base):
- Save drafts to the designated database (e.g., Scrapbox DB).
- Use structured blocks (Headings, Bullet points) rather than one giant text block.
- Apply relevant tags (e.g.,
PR原稿, ProjectName) for future retrieval.
Pitfalls & Constraints
- Character Limits: GitHub comments have limits. Split large drafts into multiple comments.
- External Approval: Flag drafts that require external (third-party) approval (e.g., partner companies) and mark them clearly as "Review Required".
- Notion Block Limits: Notion API has a 100-block limit per page creation. Split content into chunks if necessary.
Verification
