skills.rest
mastra-ai
mastra-aiOFFICIAL
Explore all skills by this author →
Website𝕏@mastraGitHub

agent-prompt-quality-bar

Audits agent system prompts against a universal quality rubric and completion criteria.

27.6k|2.7k|Updated Aug 6, 2024
Search Tags:#system prompt#fallback#Prompt Engineering#quality audit#Rubric#Agent#completion criteria
One-click install
npx skills add https://github.com/mastra-ai/mastra --skill agent-prompt-quality-bar
Or copy as Structured Prompt for Agent▼
Please help me install this Agent Skill.
Skill: agent-prompt-quality-bar
Source: https://github.com/mastra-ai/mastra/tree/main/packages/editor/src/ee/workspace/skills/agent-prompt-quality-bar
Command: npx skills add https://github.com/mastra-ai/mastra --skill agent-prompt-quality-bar

SYSTEM DOCUMENTATION & REQUIREMENTS

What problem does it solve?

Agent system prompts often fail in production because they lack completion criteria, fallback behavior for missing capabilities, and clear output formats, causing agents to stall, hallucinate success, or drift off-task.

Core Features & Use Cases

  • Run Contract Definition: Forces explicit specification of trigger, owned outcome, available capabilities, fallback paths, done criteria, and final response format before writing any prompt.
  • Nine Mandatory Sections: Enforces role, trigger, decision rules, capability awareness, missing-capability fallback, communication style, refusals, completion criteria, and a worked example in every produced prompt.
  • Final Audit Rubric: Provides an 11-point checklist to verify prompts contain no placeholders, internal jargon, unsupported capability claims, or missing refusal paths.
  • Use Case: After drafting a customer-support agent prompt with an archetype skill, run this quality bar as a final review pass to catch vague instructions like "ask the user if unsure" and replace them with explicit default behaviors.

Quick Start

Review my drafted agent system prompt against the universal quality bar and fix any sections that fail the audit rubric.

Frequently Asked Questions about agent-prompt-quality-bar

High-intent search queries and answers about installing and using this skill.

FAQPage Schema
How do I write a good system prompt for an AI agent?▼

Define a run contract first: trigger, single owned outcome, available capabilities, missing-capability fallback, done criteria, and final response format. Then write nine required sections including decision rules, refusals, completion criteria, and one worked example showing a complete run.

What sections must an agent system prompt include?▼

Nine sections are mandatory: role and outcome, trigger and input, decision rules, capability awareness, missing-capability fallback, communication style and response format, refusals and escalations, completion criteria, and a worked example demonstrating a full run.

Why does my AI agent stop mid-task or never finish?▼

The most common cause is missing completion criteria. Add an explicit numbered list of verifiable conditions, such as tool confirmation, read-back, or test output, that must all be true before the agent stops.

When should I use this quality bar instead of an archetype skill?▼

Use it when no archetype skill (coding, spreadsheet, research, support, content, ops, generic) matches the request, or as a final compile check after an archetype provides the domain-specific template.

What are common anti-patterns in agent system prompts?▼

Key anti-patterns include no completion criteria, telling the agent to ask the user when unsure, missing scope boundaries, too many attached tools, generic helpful-assistant identity, no missing-capability fallback, and untestable marketing language.

Related Skills

Explore More Software Engineering Skills→
eatbreadseatbreads
1

lark-mail

Manage Lark email workflows including drafting, replying, and attachment handling.

Community
Advanced
EmasoftEmasoft

epa-task-execution

Executes programming tasks from assignment to completion with validation against acceptance criteria.

Community
Advanced
UsernameTronUsernameTron

cc-ref-agent-archetypes

Provide 72 agent templates across 10 domains with system prompts and configurations.

Community
Advanced
SatohaCCSatohaCC

empirical-prompt-tuning

Run prompts through subagents and collect dual feedback to iteratively refine them.

Community
Advanced