Prompt Engineering Patterns#

Structured techniques for crafting prompts within agentic skills, drawn from research and the ten-pillars-agentic-skill-design framework.

Chain of Thought (CoT)#

Structure prompts to encourage step-by-step reasoning. Break complex tasks into numbered steps. Wei et al. (2022).

ReAct Pattern#

Integrate reasoning and acting in a loop (Yao et al., 2023):

Thought: [reasoning about what to do]
Action: [specific action to take]
Observation: [result of the action]
... (repeat until task complete)

Self-Reflection (Reflexion)#

Enable agents to learn from mistakes (Shinn & Labash, 2023):

• Include feedback loops in skill workflows
• Store successful patterns in memory
• Adjust strategies based on outcomes

Tree of Thoughts#

Deliberate problem solving by exploring multiple reasoning paths (Yao et al., 2023). Useful for complex architectural decisions.

Key Elements for Skill Prompts#

• System messages defining agent roles and expertise
• Stepwise instructions to guide reasoning
• Controlled temperature and token limits
• Few-shot examples for complex tasks
• Clear success/failure criteria

Anti-Patterns#

• Overly generic prompts: Vague instructions → inconsistent outputs. Use specific, structured prompts with examples.
• Ignoring token limits: Skills exceed context windows. Implement chunking and context-management.
• Unnecessary reasoning steps: LLMs often overthink. Implement trajectory monitoring and early stopping.

Composable Strategies (Fabric)#

fabric implements these patterns as composable JSON strategies applied on top of any pattern, cleanly separating what to do from how to reason:

See Also#

• ten-pillars-agentic-skill-design
• context-management
• agent-skills-standard