Agent Prompt Engineer

Write or audit AI agent system prompts component-by-component across identity, instruction architecture, behavioral constraints, tools, examples, context strategy, output format, and error handling.

What this skill does

This skill designs new agent system prompts from requirements or audits existing prompts for structural and behavioral improvements. It works through the components that determine prompt quality: persona depth, instruction placement, constraint framing, tool definition quality, example strategy, context management, output format enforcement, and error handling.

The workflow supports single-agent and multi-agent systems. In write mode it produces a complete prompt plus iteration guidance. In audit mode it scores each applicable component and returns prioritized findings with rewritten prompt text.

When to use

Use this skill for prompts such as:

• "Write a system prompt for this agent"
• "Design a prompt for an AI agent"
• "Audit this agent prompt"
• "Review my tool-use instructions"
• "Fix prompt structure for a multi-agent system"
• "Improve the context strategy in this prompt"
• "Define agent error handling and escalation behavior"
• "Tune output format instructions for reliable parsing"
• "Convert requirements into a production agent prompt"
• "Find why this agent ignores constraints or chooses tools poorly"

Install

tessl install sharaf/agent-prompt-engineer

How it's organized

Output format

Write mode returns:

# Agent System Prompt: [Name]
## First Actions
## Design Decisions
## System Prompt
## How to Iterate

Audit mode returns:

# Agent Prompt Audit: [Name]
## First Actions
## Component Scores
## Highest-Impact Fixes
## Detailed Findings
## Rewrite Blocks
## Open Questions

Audit findings cite prompt sections, assign priority, explain behavioral risk, and include concrete replacement text.

Eval results

• Registry Quality: 100%.
• Latest baseline auto-eval before the 0.1.12 targeted fixes: 97% with-context average across 3 scenarios.
• 0.1.12 adds targeted anchors for generated-prompt context strategy, iteration methodology, and audit component labels.

Built from

Migrated from the legacy flat skill /home/mamba/work/skill-builder/skills/agent-prompt-engineer.md.

Research lineage:

• Domain: agent-prompt-engineering
• Sub-domains: agent-identity-persona-design, instruction-architecture, behavioral-steering-constraints, tool-action-definitions, few-shot-examples-demonstrations, context-engineering, multi-agent-prompt-design, output-format-response-shaping, error-handling-edge-case-instructions, domain-adaptation, prompt-evaluation-iteration
• Research date: 2026-04-11