prompt logo

prompt

Free

Agent tool engineer system prompt

FreeFree tier
Inputs: textOutputs: text
Type
Open Source

About prompt

System prompt designed to instruct an AI agent to act as an 'Agent Tool Engineer'. It defines seven core responsibilities: tool selection and omission to collapse the solution space, namespacing with clear boundaries, tool prototyping and testing with real agent trajectories, context-rich returns including actionable error context, token-efficient output compression, prompt-engineering tool descriptions with imperative verbs and constraints, and agent-driven optimization loops with A/B comparisons. The prompt emphasizes that tools should be minimal, well-named, and tested to improve agent performance. Sources include Anthropic and OpenAI guidelines published in 2026.

Key Features

Tool selection and omission to collapse solution space
Namespacing with clear boundaries and no overlapping tools
Tool prototyping and testing with real agent trajectories
Context-rich returns including actionable error context
Token-efficient output compression without losing semantic content
Prompt-engineering tool descriptions with imperative verbs and explicit constraints
Agent-driven optimization loops and A/B comparisons

Pros & Cons

Pros
  • Encourages minimal and unambiguous tool interfaces
  • Provides concrete, actionable responsibilities for tool design
  • Includes principles for testing with real agent trajectories
  • Emphasizes token efficiency and context-rich returns
Cons
  • Requires integration with a compatible large language model
  • May be too abstract for beginners without prior agent experience
  • Lacks specific tool examples or code implementations

Best For

Designing effective tools for AI agentsOptimizing tool selection in agent systemsTeaching LLMs best practices for tool creation and documentationBuilding reliable agent workflows with minimal ambiguity

FAQ

What sources were used to create this prompt?
This prompt draws from Anthropic's guidelines on writing effective tools for agents and OpenAI's harness engineering, both published in 2026.