关键要点
- Frameworks are templates; choosing wrong one adds friction (extra steps, longer prompts)
- Task-first, framework-second: Identify task type (analysis, generation, code, decision), then match framework
- Standardize on 1–2 frameworks per team: CRAFT for writing, SPECS for analysis, Co-STAR for complex reasoning
- Document the choice: Add to style guide why {framework} is used for {task type}
- Review quarterly: If teams aren't using chosen framework, switch; framework adoption requires buy-in
Framework-to-Task Matching
Different tasks benefit from different frameworks; forcing the wrong one is friction.
- CRAFT (Context, Role, Action, Format, Task): Best for content writing, emails, creative work; flexible structure
- Co-STAR (Context, Objective, Style, Tone, Action, Response): Best for complex reasoning, multi-step decisions; detailed setup
- SPECS (Settings, Problem, End goal, Constraint, Success criteria): Best for analytical/technical tasks; emphasis on constraints
- Single-Step: Best for simple queries, real-time chat; no framework needed
- RTF (Role, Task, Format): Lightweight, quick; good for simple generation
Step 1: Audit Team Tasks
What does your team actually do with prompts? Categorize first.
- Content creation: Blog posts, marketing emails, social media → Use CRAFT
- Analysis/insights: Data summaries, competitive analysis, research synthesis → Use SPECS
- Decision support: Product strategy, vendor choice, architecture decision → Use Co-STAR
- Code generation: Function generation, refactoring suggestions → Use SPECS + structured output
- Customer-facing: Support responses, FAQs → Use CRAFT with brand voice overlay
CRAFT for Content Teams
Context + Role + Action + Format + Task = flexible, writer-friendly.
- Context: Background ("You are writing for startup founders learning about prompt engineering")
- Role: Persona ("You are a technical writer with 10 years experience")
- Action: Verb ("Summarize", "Draft", "Compare")
- Format: Output structure ("Bullet list", "Markdown blog post", "Email")
- Task: Specific request ("Write intro paragraph for blog about {topic}")
- Adoption: Natural for writers; minimal learning curve; easy to adjust parameters
Co-STAR for Complex Reasoning
Context + Objective + Style + Tone + Action + Response = comprehensive, detail-heavy.
- Context: Background + constraints (time, budget, dependencies)
- Objective: Clear goal (e.g., "Recommend cloud provider for startup")
- Style: Approach (comparative, analytical, creative)
- Tone: Voice (formal, casual, friendly)
- Action: Specific steps (list options, evaluate criteria, rank)
- Response: Output format (decision matrix, recommendation with justification)
- Adoption: Takes practice; best for complex multi-step decisions
SPECS for Technical Tasks
Settings + Problem + End goal + Constraint + Success = constraint-driven, precise.
- Settings: Context (data, tools, environment)
- Problem: What needs solving (e.g., "Slow SQL query")
- End goal: Desired outcome (fast query, readable code)
- Constraint: Limits (database schema, library versions)
- Success criteria: How to measure (query <1s, test coverage >80%)
- Adoption: Natural for engineers; structured, logical progression
Adoption Strategy
Choosing a framework is useless if the team doesn't use it.
- Training: 30-min workshop per framework; show 3 examples; practice on real task
- Documentation: Add to style guide with examples; link from Slack bot
- Enforcement: Code review comment if framework not used; gentle, not punitive
- Feedback loop: "What framework works best for your task?" Survey quarterly
- Flexibility: Allow variations; framework is template, not rule
Common Mistakes
- Picking framework for the wrong reason (it's trendy, not task-fit)
- Forcing one framework for all tasks (like forcing Python for DevOps, shell scripting)
- Not documenting why chosen (team doesn't understand; compliance audits fail)
- Adopting framework without training (team ignores it; doesn't know how)
- Never revisiting choice (framework doesn't fit; team works around it instead of switching)
Sources
- PromptQuorum framework comparison guide, 2026
- CRAFT, Co-STAR, SPECS original research papers
- Team surveys from 10+ organizations on framework adoption