system-prompt-engineering

What it does

This skill enables AI agents to automatically refine and optimize their own system prompts based on performance feedback. It creates a self-improving loop where the agent analyzes its outputs and iteratively adjusts instructions to better align with user goals.

When to use it

• Iterative Development: When building complex workflows that require the agent to learn from its mistakes over time.
• Persona Tuning: To dynamically adjust the agent's tone, expertise level, or specific constraints without manual intervention.
• Error Reduction: When an agent consistently fails at a specific task type and needs immediate instruction updates to correct behavior.
• Context Adaptation: For scenarios where user requirements shift frequently, allowing the prompt to evolve alongside changing needs.

Key capabilities

• Automatically generates optimized system prompts based on interaction history.
• Analyzes output quality to identify areas for instruction improvement.
• Maintains a versioned history of prompt iterations for comparison.
• Integrates seamlessly with Claude Code plugins for real-time updates.

Example prompts

• "Analyze my last three failed code generation attempts and rewrite the system prompt to prevent these specific errors."
• "Adjust the system instructions to adopt a more senior developer persona while maintaining strict security protocols."
• "Generate an updated system prompt that incorporates the new API documentation I just uploaded."

Tips & gotchas

This skill requires a baseline of successful interactions to effectively analyze what works and what doesn't; avoid using it immediately on brand-new tasks with no history. Ensure you have clear success criteria defined, as the optimization logic relies heavily on distinguishing between good and bad outputs.